{"id":6300,"date":"2026-10-22T13:16:11","date_gmt":"2026-10-22T16:16:11","guid":{"rendered":"https:\/\/tecnologiaunica.com.br\/?p=6300"},"modified":"2026-10-23T12:54:56","modified_gmt":"2026-10-23T15:54:56","slug":"explicabilidade-de-ia-em-seguros","status":"publish","type":"post","link":"https:\/\/tecnologiaunica.com.br\/it\/blog\/explicabilidade-de-ia-em-seguros\/","title":{"rendered":"Explicabilidade de IA em seguros: como sustentar cada decis\u00e3o"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">Explicabilidade de IA em seguros \u00e9 a capacidade de apresentar raz\u00f5es compreens\u00edveis e fi\u00e9is para os resultados de um sistema que participa de decis\u00f5es sobre aceita\u00e7\u00e3o, pre\u00e7o ou sinistros. Para sustentar uma decis\u00e3o espec\u00edfica, essa capacidade precisa estar acompanhada de fundamenta\u00e7\u00e3o adequada e de registros que permitam verificar o que aconteceu.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A diferen\u00e7a aparece quando um segurado contesta uma recusa meses depois. A seguradora precisa recuperar os dados considerados, identificar o modelo e as regras utilizados, demonstrar como o resultado influenciou a decis\u00e3o e apresentar seus fundamentos.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Guardar tudo isso ajuda a reconstituir o caso. Ainda ser\u00e1 necess\u00e1rio avaliar se a explica\u00e7\u00e3o corresponde ao funcionamento do sistema e se a decis\u00e3o \u00e9 defens\u00e1vel.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Para uma vis\u00e3o mais ampla dos usos da tecnologia no setor, veja o panorama de <strong><a href=\"https:\/\/tecnologiaunica.com.br\/blog\/inteligencia-artificial-em-seguros\/\" target=\"_blank\" rel=\"noreferrer noopener\">aplica\u00e7\u00f5es e riscos de intelig\u00eancia artificial em seguros<\/a><\/strong>. Aqui, o foco \u00e9 o que uma decis\u00e3o apoiada por IA precisa deixar dispon\u00edvel para ser compreendida, revisada e sustentada.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Explica\u00e7\u00e3o, justificativa e rastro cumprem fun\u00e7\u00f5es diferentes<\/h2>\n\n\n\n<style>\n  .tu-ai-explanation-justification-trace,\n  .tu-ai-explanation-justification-trace * {\n    box-sizing: border-box;\n  }\n\n  .tu-ai-explanation-justification-trace {\n    --tu-red: #DD4E4B;\n    --tu-black: #000000;\n    --tu-white: #FFFFFF;\n    --tu-soft: #FFF7F7;\n    --tu-line: #E7E7E7;\n    --tu-text: #383838;\n\n    width: 100%;\n    max-width: 100%;\n    margin: 24px 0;\n\n    font-family: \"DM Sans\", Arial, sans-serif;\n  }\n\n  .tu-ai-explanation-justification-trace__grid {\n    display: grid;\n    grid-template-columns: repeat(3, minmax(0, 1fr));\n    gap: 14px;\n\n    width: 100%;\n  }\n\n  .tu-ai-explanation-justification-trace__card {\n    min-width: 0;\n    padding: 20px 18px;\n\n    background: var(--tu-white);\n    border: 1px solid var(--tu-line);\n    border-top: 4px solid var(--tu-red);\n    border-radius: 4px 15px 15px 15px;\n  }\n\n  .tu-ai-explanation-justification-trace__card:nth-child(2) {\n    background: var(--tu-soft);\n  }\n\n  .tu-ai-explanation-justification-trace__card--highlight {\n    background: var(--tu-red) !important;\n    border: 2px solid var(--tu-black);\n    border-top-width: 4px;\n  }\n\n  .tu-ai-explanation-justification-trace__header {\n    display: flex;\n    align-items: center;\n    gap: 11px;\n\n    margin-bottom: 11px;\n  }\n\n  .tu-ai-explanation-justification-trace__number {\n    display: flex;\n    align-items: center;\n    justify-content: center;\n\n    flex: 0 0 30px;\n    width: 30px;\n    height: 30px;\n\n    background: var(--tu-red);\n    border-radius: 50%;\n\n    color: var(--tu-white);\n    font-size: 9.5px;\n    font-weight: 700;\n    line-height: 1;\n  }\n\n  .tu-ai-explanation-justification-trace__card--highlight\n  .tu-ai-explanation-justification-trace__number {\n    background: var(--tu-white);\n    color: var(--tu-red);\n  }\n\n  .tu-ai-explanation-justification-trace__title {\n    margin: 0 !important;\n    padding: 0 !important;\n\n    color: var(--tu-black);\n    font-size: 14.5px;\n    font-weight: 700;\n    line-height: 1.4;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-explanation-justification-trace__text {\n    margin: 0 !important;\n    padding: 0 !important;\n\n    color: var(--tu-text) !important;\n    font-size: 12.8px;\n    font-weight: 400;\n    line-height: 1.62;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-explanation-justification-trace__text + .tu-ai-explanation-justification-trace__text {\n    margin-top: 9px !important;\n  }\n\n  .tu-ai-explanation-justification-trace__text strong {\n    color: var(--tu-black);\n    font-weight: 700;\n  }\n\n  .tu-ai-explanation-justification-trace__card--highlight\n  .tu-ai-explanation-justification-trace__title,\n  .tu-ai-explanation-justification-trace__card--highlight\n  .tu-ai-explanation-justification-trace__text,\n  .tu-ai-explanation-justification-trace__card--highlight\n  .tu-ai-explanation-justification-trace__text strong {\n    color: var(--tu-white) !important;\n  }\n\n  .tu-ai-explanation-justification-trace__distinction {\n    display: block;\n\n    margin-top: 13px;\n    padding-top: 11px;\n\n    border-top: 1px solid var(--tu-line);\n\n    color: var(--tu-black);\n    font-size: 12.2px;\n    font-weight: 700;\n    line-height: 1.5;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-explanation-justification-trace__distinction::before {\n    content: \"\u2192\";\n    margin-right: 7px;\n\n    color: var(--tu-red);\n    font-size: 14px;\n    font-weight: 700;\n  }\n\n  .tu-ai-explanation-justification-trace__card--highlight\n  .tu-ai-explanation-justification-trace__distinction {\n    color: var(--tu-white);\n    border-top-color: rgba(255, 255, 255, .28);\n  }\n\n  .tu-ai-explanation-justification-trace__card--highlight\n  .tu-ai-explanation-justification-trace__distinction::before {\n    color: var(--tu-white);\n  }\n\n  @media (max-width: 920px) {\n    .tu-ai-explanation-justification-trace__grid {\n      grid-template-columns: 1fr;\n    }\n  }\n\n  @media (max-width: 620px) {\n    .tu-ai-explanation-justification-trace {\n      margin: 20px 0;\n    }\n\n    .tu-ai-explanation-justification-trace__grid {\n      gap: 10px;\n    }\n\n    .tu-ai-explanation-justification-trace__card {\n      padding: 17px 15px;\n\n      border-radius: 4px 13px 13px 13px;\n    }\n\n    .tu-ai-explanation-justification-trace__number {\n      flex-basis: 28px;\n      width: 28px;\n      height: 28px;\n\n      font-size: 9px;\n    }\n\n    .tu-ai-explanation-justification-trace__title {\n      font-size: 14px;\n    }\n\n    .tu-ai-explanation-justification-trace__text {\n      font-size: 12.55px;\n    }\n\n    .tu-ai-explanation-justification-trace__distinction {\n      font-size: 12.05px;\n    }\n  }\n<\/style>\n\n<div class=\"tu-ai-explanation-justification-trace\">\n\n  <div class=\"tu-ai-explanation-justification-trace__grid\">\n\n    <section class=\"tu-ai-explanation-justification-trace__card\">\n\n      <div class=\"tu-ai-explanation-justification-trace__header\">\n        <span class=\"tu-ai-explanation-justification-trace__number\" aria-hidden=\"true\">01<\/span>\n\n        <p class=\"tu-ai-explanation-justification-trace__title\">\n          Explica\u00e7\u00e3o\n        <\/p>\n      <\/div>\n\n      <p class=\"tu-ai-explanation-justification-trace__text\">\n        Descreve <strong>como o sistema produziu determinado resultado<\/strong>.\n        Pode apresentar uma regra percorrida, a contribui\u00e7\u00e3o de vari\u00e1veis ou os\n        elementos de um documento que sustentaram uma extra\u00e7\u00e3o.\n      <\/p>\n\n      <span class=\"tu-ai-explanation-justification-trace__distinction\">\n        O formato depende da t\u00e9cnica utilizada e de quem precisa compreender o resultado.\n      <\/span>\n\n    <\/section>\n\n\n    <section class=\"tu-ai-explanation-justification-trace__card\">\n\n      <div class=\"tu-ai-explanation-justification-trace__header\">\n        <span class=\"tu-ai-explanation-justification-trace__number\" aria-hidden=\"true\">02<\/span>\n\n        <p class=\"tu-ai-explanation-justification-trace__title\">\n          Justificativa\n        <\/p>\n      <\/div>\n\n      <p class=\"tu-ai-explanation-justification-trace__text\">\n        Demonstra <strong>por que a decis\u00e3o \u00e9 defens\u00e1vel diante dos fatos e das\n        regras aplic\u00e1veis<\/strong>.\n      <\/p>\n\n      <p class=\"tu-ai-explanation-justification-trace__text\">\n        Explicar que uma vari\u00e1vel influenciou o resultado n\u00e3o comprova que seu uso\n        foi adequado nem que havia fundamento para recusar uma proposta ou uma cobertura.\n      <\/p>\n\n      <span class=\"tu-ai-explanation-justification-trace__distinction\">\n        Explicar o resultado e sustentar a decis\u00e3o s\u00e3o exig\u00eancias diferentes.\n      <\/span>\n\n    <\/section>\n\n\n    <section class=\"tu-ai-explanation-justification-trace__card tu-ai-explanation-justification-trace__card--highlight\">\n\n      <div class=\"tu-ai-explanation-justification-trace__header\">\n        <span class=\"tu-ai-explanation-justification-trace__number\" aria-hidden=\"true\">03<\/span>\n\n        <p class=\"tu-ai-explanation-justification-trace__title\">\n          Rastro\n        <\/p>\n      <\/div>\n\n      <p class=\"tu-ai-explanation-justification-trace__text\">\n        Re\u00fane as <strong>evid\u00eancias da execu\u00e7\u00e3o<\/strong>: dados utilizados,\n        vers\u00f5es, resultados, interven\u00e7\u00f5es humanas e comunica\u00e7\u00e3o realizada.\n      <\/p>\n\n      <span class=\"tu-ai-explanation-justification-trace__distinction\">\n        \u00c9 o registro que permite reconstruir o que aconteceu ao longo da decis\u00e3o.\n      <\/span>\n\n    <\/section>\n\n  <\/div>\n\n<\/div>\n\n\n\n<p class=\"wp-block-paragraph\">O projeto precisa tratar essas tr\u00eas necessidades em conjunto, preservando suas diferen\u00e7as.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Quem precisa entender a decis\u00e3o?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">A mesma informa\u00e7\u00e3o n\u00e3o atende a todos os destinat\u00e1rios.<\/p>\n\n\n\n<style>\n  .tu-ai-audiences-table,\n  .tu-ai-audiences-table * {\n    box-sizing: border-box;\n  }\n\n  .tu-ai-audiences-table {\n    --tu-red: #DD4E4B;\n    --tu-black: #000000;\n    --tu-white: #FFFFFF;\n    --tu-soft: #FFF7F7;\n    --tu-line: #E7E7E7;\n    --tu-text: #333333;\n\n    width: 100%;\n    max-width: 100%;\n    margin: 24px 0;\n\n    font-family: \"DM Sans\", Arial, sans-serif;\n  }\n\n  .tu-ai-audiences-table__container {\n    width: 100%;\n    max-width: 100%;\n    overflow: hidden;\n\n    background: var(--tu-white);\n    border: 1px solid #D9D9D9;\n    border-radius: 4px 16px 16px 16px;\n  }\n\n  .tu-ai-audiences-table table {\n    width: 100% !important;\n    max-width: 100% !important;\n    margin: 0 !important;\n\n    border-collapse: separate;\n    border-spacing: 0;\n    table-layout: fixed;\n  }\n\n  .tu-ai-audiences-table th,\n  .tu-ai-audiences-table td {\n    padding: 14px 15px;\n\n    vertical-align: top;\n    text-align: left;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-audiences-table thead th {\n    background: var(--tu-red);\n    color: var(--tu-black);\n\n    border-right: 1px solid rgba(0, 0, 0, .15);\n\n    font-size: 11.9px;\n    font-weight: 700;\n    line-height: 1.4;\n  }\n\n  .tu-ai-audiences-table thead th:last-child {\n    border-right: 0;\n  }\n\n  .tu-ai-audiences-table thead th:nth-child(1) {\n    width: 26%;\n  }\n\n  .tu-ai-audiences-table thead th:nth-child(2) {\n    width: 38%;\n  }\n\n  .tu-ai-audiences-table thead th:nth-child(3) {\n    width: 36%;\n  }\n\n  .tu-ai-audiences-table tbody td {\n    position: relative;\n\n    background: var(--tu-white);\n\n    border-right: 1px solid var(--tu-line);\n    border-bottom: 1px solid var(--tu-line);\n\n    color: var(--tu-text);\n\n    font-size: 12.3px;\n    font-weight: 400;\n    line-height: 1.55;\n  }\n\n  .tu-ai-audiences-table tbody tr:nth-child(even) td {\n    background: var(--tu-soft);\n  }\n\n  .tu-ai-audiences-table tbody td:last-child {\n    border-right: 0;\n  }\n\n  .tu-ai-audiences-table tbody tr:last-child td {\n    border-bottom: 0;\n  }\n\n  .tu-ai-audiences-table tbody td:first-child {\n    padding-left: 19px;\n\n    color: var(--tu-black);\n    font-weight: 700;\n  }\n\n  .tu-ai-audiences-table tbody td:first-child::before {\n    content: \"\";\n\n    position: absolute;\n    top: 0;\n    bottom: 0;\n    left: 0;\n\n    width: 4px;\n\n    background: var(--tu-red);\n  }\n\n  .tu-ai-audiences-table__delivery {\n    color: var(--tu-black) !important;\n    font-weight: 600 !important;\n  }\n\n  @media (max-width: 680px) {\n    .tu-ai-audiences-table {\n      margin: 20px 0;\n    }\n\n    .tu-ai-audiences-table__container {\n      overflow: visible;\n\n      background: transparent;\n      border: 0;\n      border-radius: 0;\n    }\n\n    .tu-ai-audiences-table table,\n    .tu-ai-audiences-table tbody,\n    .tu-ai-audiences-table tr,\n    .tu-ai-audiences-table td {\n      display: block;\n\n      width: 100% !important;\n      max-width: 100% !important;\n    }\n\n    .tu-ai-audiences-table thead {\n      position: absolute;\n\n      width: 1px;\n      height: 1px;\n\n      margin: -1px;\n      padding: 0;\n\n      overflow: hidden;\n\n      clip: rect(0, 0, 0, 0);\n      white-space: nowrap;\n\n      border: 0;\n    }\n\n    .tu-ai-audiences-table tbody {\n      display: grid;\n      gap: 12px;\n    }\n\n    .tu-ai-audiences-table tbody tr {\n      overflow: hidden;\n\n      background: var(--tu-white);\n      border: 1px solid #DCDCDC;\n      border-radius: 4px 13px 13px 13px;\n    }\n\n    .tu-ai-audiences-table tbody td,\n    .tu-ai-audiences-table tbody tr:nth-child(even) td {\n      padding: 11px 14px;\n\n      background: var(--tu-white);\n\n      border-right: 0;\n      border-bottom: 1px solid var(--tu-line);\n\n      font-size: 12.7px;\n      line-height: 1.5;\n    }\n\n    .tu-ai-audiences-table tbody td:last-child {\n      border-bottom: 0;\n    }\n\n    .tu-ai-audiences-table tbody td::before {\n      display: block;\n\n      margin-bottom: 4px;\n\n      color: #6A6A6A;\n\n      font-size: 9.8px;\n      font-weight: 700;\n      line-height: 1.2;\n      letter-spacing: .035em;\n      text-transform: uppercase;\n\n      content: attr(data-label);\n    }\n\n    .tu-ai-audiences-table tbody td:first-child {\n      padding: 13px 14px;\n\n      background: var(--tu-red) !important;\n      color: var(--tu-black) !important;\n\n      font-size: 13.3px;\n      font-weight: 700;\n    }\n\n    .tu-ai-audiences-table tbody td:first-child::before {\n      position: static;\n\n      display: block;\n\n      width: auto;\n      height: auto;\n      margin-bottom: 4px;\n\n      background: none;\n\n      color: rgba(0, 0, 0, .65);\n\n      content: attr(data-label);\n    }\n  }\n<\/style>\n\n<div class=\"tu-ai-audiences-table\">\n\n  <div class=\"tu-ai-audiences-table__container\">\n\n    <table>\n\n      <thead>\n        <tr>\n          <th scope=\"col\">Destinat\u00e1rio<\/th>\n          <th scope=\"col\">O que precisa compreender<\/th>\n          <th scope=\"col\">Entrega adequada<\/th>\n        <\/tr>\n      <\/thead>\n\n      <tbody>\n\n        <tr>\n          <td data-label=\"Destinat\u00e1rio\">\n            Segurado ou proponente\n          <\/td>\n\n          <td data-label=\"O que precisa compreender\">\n            Motivo da decis\u00e3o, fatos considerados e caminho para questionamento.\n          <\/td>\n\n          <td\n            class=\"tu-ai-audiences-table__delivery\"\n            data-label=\"Entrega adequada\"\n          >\n            Comunica\u00e7\u00e3o clara, relacionada ao caso.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Destinat\u00e1rio\">\n            Analista ou revisor\n          <\/td>\n\n          <td data-label=\"O que precisa compreender\">\n            Fundamentos da recomenda\u00e7\u00e3o, limita\u00e7\u00f5es e elementos que podem alter\u00e1-la.\n          <\/td>\n\n          <td\n            class=\"tu-ai-audiences-table__delivery\"\n            data-label=\"Entrega adequada\"\n          >\n            Informa\u00e7\u00f5es acess\u00edveis na tela, com documentos e crit\u00e9rios relevantes.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Destinat\u00e1rio\">\n            Auditoria e autoridades competentes\n          <\/td>\n\n          <td data-label=\"O que precisa compreender\">\n            Como a decis\u00e3o foi produzida e quais controles foram aplicados.\n          <\/td>\n\n          <td\n            class=\"tu-ai-audiences-table__delivery\"\n            data-label=\"Entrega adequada\"\n          >\n            Evid\u00eancias recuper\u00e1veis, com hist\u00f3rico e integridade.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Destinat\u00e1rio\">\n            Equipes t\u00e9cnica e de riscos\n          <\/td>\n\n          <td data-label=\"O que precisa compreender\">\n            Comportamento do modelo, falhas e mudan\u00e7as ao longo do tempo.\n          <\/td>\n\n          <td\n            class=\"tu-ai-audiences-table__delivery\"\n            data-label=\"Entrega adequada\"\n          >\n            Avalia\u00e7\u00f5es de desempenho, estabilidade e resultados por grupos pertinentes.\n          <\/td>\n        <\/tr>\n\n      <\/tbody>\n\n    <\/table>\n\n  <\/div>\n\n<\/div>\n\n\n\n<p class=\"wp-block-paragraph\">O NIST prop\u00f5e quatro princ\u00edpios \u00fateis: o sistema deve oferecer explica\u00e7\u00f5es; elas precisam fazer sentido para o destinat\u00e1rio; devem refletir corretamente o processo; e o sistema deve reconhecer seus limites de conhecimento.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Isso exige avaliar tanto a fidelidade t\u00e9cnica quanto a compreens\u00e3o. Uma explica\u00e7\u00e3o pode estar matematicamente correta e ser in\u00fatil para quem precisa decidir se concorda com ela.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">O que as obriga\u00e7\u00f5es brasileiras j\u00e1 exigem<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">O enquadramento depende do tipo de decis\u00e3o, dos dados tratados e das regras aplic\u00e1veis ao produto. H\u00e1 deveres distintos que precisam ser traduzidos em controles operacionais.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">LGPD: crit\u00e9rios, procedimentos e revis\u00e3o<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">O art. 20 da LGPD assegura ao titular o direito de solicitar revis\u00e3o de decis\u00f5es tomadas unicamente com base em tratamento automatizado de dados pessoais que afetem seus interesses. O \u00a7 1\u00ba exige, quando solicitadas, informa\u00e7\u00f5es claras e adequadas sobre crit\u00e9rios e procedimentos, observados os segredos comercial e industrial.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A reda\u00e7\u00e3o vigente n\u00e3o determina expressamente que a revis\u00e3o seja feita por pessoa natural. Quando a opera\u00e7\u00e3o adota supervis\u00e3o humana como controle, por\u00e9m, precisa verificar se o revisor disp\u00f5e de informa\u00e7\u00e3o e autoridade para intervir.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Lei do Contrato de Seguro: fundamenta\u00e7\u00e3o das recusas<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">A Lei n\u00ba 15.040\/2024 estabelece obriga\u00e7\u00f5es diretamente relacionadas ao tema:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Propostas:<\/strong> o art. 49 exige comunica\u00e7\u00e3o da justificativa da recusa. O prazo m\u00e1ximo \u00e9 de 25 dias, com rein\u00edcio nas hip\u00f3teses do \u00a7 2\u00ba.<\/li>\n\n\n\n<li><strong>Subscri\u00e7\u00e3o:<\/strong> o art. 51 veda pol\u00edticas t\u00e9cnicas e comerciais conducentes \u00e0 discrimina\u00e7\u00e3o social ou prejudiciais \u00e0 livre iniciativa empresarial.<\/li>\n\n\n\n<li><strong>Sinistros:<\/strong> o art. 83 determina a entrega dos documentos que fundamentem a negativa, ressalvadas as exce\u00e7\u00f5es legais de sigilo e prote\u00e7\u00e3o de terceiros.<\/li>\n\n\n\n<li><strong>Recusa de cobertura:<\/strong> o art. 86, \u00a7 6\u00ba, exige manifesta\u00e7\u00e3o expressa e motivada e restringe a altera\u00e7\u00e3o posterior do fundamento, salvo conhecimento de fatos antes desconhecidos.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Esses dispositivos tornam relevante preservar a rela\u00e7\u00e3o entre evid\u00eancias, crit\u00e9rios e decis\u00e3o.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Regula\u00e7\u00e3o setorial: controles e regras do contrato<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">A Resolu\u00e7\u00e3o CNSP n\u00ba 416\/2021 trata de controles internos, gest\u00e3o de riscos e auditoria interna. Processos apoiados por modelos devem ser avaliados dentro dessa estrutura, conforme os riscos envolvidos. A norma continua entre as refer\u00eancias utilizadas pela pr\u00f3pria SUSEP em suas orienta\u00e7\u00f5es de seguran\u00e7a para supervisionadas.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Nos seguros de danos, a Resolu\u00e7\u00e3o CNSP n\u00ba 496\/2026 tamb\u00e9m precisa entrar no mapeamento. Ela disciplina contratos e procedimentos de sinistro, com adapta\u00e7\u00e3o dos planos existentes at\u00e9 4 de janeiro de 2027 e aplica\u00e7\u00e3o obrigat\u00f3ria aos contratos formados ou renovados a partir de 5 de janeiro de 2027, observadas suas disposi\u00e7\u00f5es de transi\u00e7\u00e3o. <\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Essas refer\u00eancias n\u00e3o estabelecem uma t\u00e9cnica universal de explica\u00e7\u00e3o. A aplica\u00e7\u00e3o pr\u00e1tica exige identificar o que cada decis\u00e3o demanda e demonstrar como o sistema atende a essa necessidade.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Modelos interpret\u00e1veis ou explica\u00e7\u00e3o p\u00f3s-hoc?<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">A escolha deve come\u00e7ar pelo problema e pelo desempenho necess\u00e1rio. Modelos interpret\u00e1veis podem reduzir o esfor\u00e7o de explica\u00e7\u00e3o, especialmente em dados estruturados, mas continuam exigindo valida\u00e7\u00e3o, documenta\u00e7\u00e3o e comunica\u00e7\u00e3o adequada.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">O trabalho de Cynthia Rudin sobre decis\u00f5es de alto impacto apresenta argumentos para considerar modelos interpret\u00e1veis desde o desenho da solu\u00e7\u00e3o, em vez de assumir que a complexidade \u00e9 inevit\u00e1vel.<\/p>\n\n\n\n<style>\n  .tu-ai-methods-explainability-table,\n  .tu-ai-methods-explainability-table * {\n    box-sizing: border-box;\n  }\n\n  .tu-ai-methods-explainability-table {\n    --tu-red: #DD4E4B;\n    --tu-black: #000000;\n    --tu-white: #FFFFFF;\n    --tu-soft: #FFF7F7;\n    --tu-line: #E7E7E7;\n    --tu-text: #333333;\n\n    width: 100%;\n    max-width: 100%;\n    margin: 24px 0;\n\n    font-family: \"DM Sans\", Arial, sans-serif;\n  }\n\n  .tu-ai-methods-explainability-table__container {\n    width: 100%;\n    max-width: 100%;\n    overflow: hidden;\n\n    background: var(--tu-white);\n    border: 1px solid #D9D9D9;\n    border-radius: 4px 16px 16px 16px;\n  }\n\n  .tu-ai-methods-explainability-table table {\n    width: 100% !important;\n    max-width: 100% !important;\n    margin: 0 !important;\n\n    border-collapse: separate;\n    border-spacing: 0;\n    table-layout: fixed;\n  }\n\n  .tu-ai-methods-explainability-table th,\n  .tu-ai-methods-explainability-table td {\n    padding: 14px 15px;\n\n    vertical-align: top;\n    text-align: left;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-methods-explainability-table thead th {\n    background: var(--tu-red);\n    color: var(--tu-black);\n\n    border-right: 1px solid rgba(0, 0, 0, .15);\n\n    font-size: 11.9px;\n    font-weight: 700;\n    line-height: 1.4;\n  }\n\n  .tu-ai-methods-explainability-table thead th:last-child {\n    border-right: 0;\n  }\n\n  .tu-ai-methods-explainability-table thead th:nth-child(1) {\n    width: 27%;\n  }\n\n  .tu-ai-methods-explainability-table thead th:nth-child(2) {\n    width: 35%;\n  }\n\n  .tu-ai-methods-explainability-table thead th:nth-child(3) {\n    width: 38%;\n  }\n\n  .tu-ai-methods-explainability-table tbody td {\n    position: relative;\n\n    background: var(--tu-white);\n\n    border-right: 1px solid var(--tu-line);\n    border-bottom: 1px solid var(--tu-line);\n\n    color: var(--tu-text);\n\n    font-size: 12.25px;\n    font-weight: 400;\n    line-height: 1.55;\n  }\n\n  .tu-ai-methods-explainability-table tbody tr:nth-child(even) td {\n    background: var(--tu-soft);\n  }\n\n  .tu-ai-methods-explainability-table tbody td:last-child {\n    border-right: 0;\n  }\n\n  .tu-ai-methods-explainability-table tbody tr:last-child td {\n    border-bottom: 0;\n  }\n\n  .tu-ai-methods-explainability-table tbody td:first-child {\n    padding-left: 19px;\n\n    color: var(--tu-black);\n    font-weight: 700;\n  }\n\n  .tu-ai-methods-explainability-table tbody td:first-child::before {\n    content: \"\";\n\n    position: absolute;\n    top: 0;\n    bottom: 0;\n    left: 0;\n\n    width: 4px;\n\n    background: var(--tu-red);\n  }\n\n  .tu-ai-methods-explainability-table__care {\n    color: var(--tu-black) !important;\n    font-weight: 600 !important;\n  }\n\n  @media (max-width: 680px) {\n    .tu-ai-methods-explainability-table {\n      margin: 20px 0;\n    }\n\n    .tu-ai-methods-explainability-table__container {\n      overflow: visible;\n\n      background: transparent;\n      border: 0;\n      border-radius: 0;\n    }\n\n    .tu-ai-methods-explainability-table table,\n    .tu-ai-methods-explainability-table tbody,\n    .tu-ai-methods-explainability-table tr,\n    .tu-ai-methods-explainability-table td {\n      display: block;\n\n      width: 100% !important;\n      max-width: 100% !important;\n    }\n\n    .tu-ai-methods-explainability-table thead {\n      position: absolute;\n\n      width: 1px;\n      height: 1px;\n\n      margin: -1px;\n      padding: 0;\n\n      overflow: hidden;\n\n      clip: rect(0, 0, 0, 0);\n      white-space: nowrap;\n\n      border: 0;\n    }\n\n    .tu-ai-methods-explainability-table tbody {\n      display: grid;\n      gap: 12px;\n    }\n\n    .tu-ai-methods-explainability-table tbody tr {\n      overflow: hidden;\n\n      background: var(--tu-white);\n      border: 1px solid #DCDCDC;\n      border-radius: 4px 13px 13px 13px;\n    }\n\n    .tu-ai-methods-explainability-table tbody td,\n    .tu-ai-methods-explainability-table tbody tr:nth-child(even) td {\n      padding: 11px 14px;\n\n      background: var(--tu-white);\n\n      border-right: 0;\n      border-bottom: 1px solid var(--tu-line);\n\n      font-size: 12.7px;\n      line-height: 1.5;\n    }\n\n    .tu-ai-methods-explainability-table tbody td:last-child {\n      border-bottom: 0;\n    }\n\n    .tu-ai-methods-explainability-table tbody td::before {\n      display: block;\n\n      margin-bottom: 4px;\n\n      color: #6A6A6A;\n\n      font-size: 9.8px;\n      font-weight: 700;\n      line-height: 1.2;\n      letter-spacing: .035em;\n      text-transform: uppercase;\n\n      content: attr(data-label);\n    }\n\n    .tu-ai-methods-explainability-table tbody td:first-child {\n      padding: 13px 14px;\n\n      background: var(--tu-red) !important;\n      color: var(--tu-black) !important;\n\n      font-size: 13.3px;\n      font-weight: 700;\n    }\n\n    .tu-ai-methods-explainability-table tbody td:first-child::before {\n      position: static;\n\n      display: block;\n\n      width: auto;\n      height: auto;\n      margin-bottom: 4px;\n\n      background: none;\n\n      color: rgba(0, 0, 0, .65);\n\n      content: attr(data-label);\n    }\n  }\n<\/style>\n\n<div class=\"tu-ai-methods-explainability-table\">\n\n  <div class=\"tu-ai-methods-explainability-table__container\">\n\n    <table>\n\n      <thead>\n        <tr>\n          <th scope=\"col\">Abordagem<\/th>\n          <th scope=\"col\">Contribui\u00e7\u00e3o para a explica\u00e7\u00e3o<\/th>\n          <th scope=\"col\">Cuidado necess\u00e1rio<\/th>\n        <\/tr>\n      <\/thead>\n\n      <tbody>\n\n        <tr>\n          <td data-label=\"Abordagem\">\n            Regras expl\u00edcitas\n          <\/td>\n\n          <td data-label=\"Contribui\u00e7\u00e3o para a explica\u00e7\u00e3o\">\n            Permitem identificar condi\u00e7\u00f5es e caminhos aplicados.\n          <\/td>\n\n          <td\n            class=\"tu-ai-methods-explainability-table__care\"\n            data-label=\"Cuidado necess\u00e1rio\"\n          >\n            Regras numerosas ou conflitantes podem dificultar a compreens\u00e3o.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Abordagem\">\n            Modelos lineares, GLMs e scorecards\n          <\/td>\n\n          <td data-label=\"Contribui\u00e7\u00e3o para a explica\u00e7\u00e3o\">\n            Podem oferecer contribui\u00e7\u00f5es leg\u00edveis, conforme sua constru\u00e7\u00e3o.\n          <\/td>\n\n          <td\n            class=\"tu-ai-methods-explainability-table__care\"\n            data-label=\"Cuidado necess\u00e1rio\"\n          >\n            Transforma\u00e7\u00f5es, intera\u00e7\u00f5es e grande quantidade de vari\u00e1veis podem reduzir a interpretabilidade.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Abordagem\">\n            GAMs\n          <\/td>\n\n          <td data-label=\"Contribui\u00e7\u00e3o para a explica\u00e7\u00e3o\">\n            Permitem examinar efeitos aditivos, inclusive n\u00e3o lineares.\n          <\/td>\n\n          <td\n            class=\"tu-ai-methods-explainability-table__care\"\n            data-label=\"Cuidado necess\u00e1rio\"\n          >\n            Os efeitos podem variar ao longo dos valores da vari\u00e1vel; n\u00e3o h\u00e1 necessariamente um \u00fanico peso ou sinal.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Abordagem\">\n            Ensembles com atribui\u00e7\u00e3o p\u00f3s-hoc\n          <\/td>\n\n          <td data-label=\"Contribui\u00e7\u00e3o para a explica\u00e7\u00e3o\">\n            Ajudam a analisar a contribui\u00e7\u00e3o das vari\u00e1veis para uma sa\u00edda.\n          <\/td>\n\n          <td\n            class=\"tu-ai-methods-explainability-table__care\"\n            data-label=\"Cuidado necess\u00e1rio\"\n          >\n            O resultado depende do m\u00e9todo, de suas hip\u00f3teses e refer\u00eancias.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Abordagem\">\n            LLMs sobre documentos\n          <\/td>\n\n          <td data-label=\"Contribui\u00e7\u00e3o para a explica\u00e7\u00e3o\">\n            Podem extrair informa\u00e7\u00f5es e apresentar trechos de apoio.\n          <\/td>\n\n          <td\n            class=\"tu-ai-methods-explainability-table__care\"\n            data-label=\"Cuidado necess\u00e1rio\"\n          >\n            Cita\u00e7\u00f5es e respostas precisam ser verificadas no documento original.\n          <\/td>\n        <\/tr>\n\n      <\/tbody>\n\n    <\/table>\n\n  <\/div>\n\n<\/div>\n\n\n\n<p class=\"wp-block-paragraph\">Os GAMs acomodam rela\u00e7\u00f5es n\u00e3o lineares mantendo uma estrutura aditiva, como descreve a documenta\u00e7\u00e3o do pyGAM.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Tamb\u00e9m \u00e9 impreciso tratar toda atribui\u00e7\u00e3o p\u00f3s-hoc como uma aproxima\u00e7\u00e3o num\u00e9rica. O Tree SHAP, por exemplo, oferece c\u00e1lculos exatos sob determinadas hip\u00f3teses. Essa exatid\u00e3o n\u00e3o estabelece causalidade nem demonstra a legitimidade da decis\u00e3o. <\/p>\n\n\n\n<h3 class=\"wp-block-heading\">O limite da regra aplicada sobre um score<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Uma regra como \u201cencaminhar para an\u00e1lise quando o score superar determinado valor\u201d explicita o encaminhamento. Ela n\u00e3o esclarece, sozinha, por que aquele caso recebeu o score.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Quanto mais o resultado do modelo determinar pre\u00e7o, recusa ou prioridade de atendimento, maior a necessidade de compreender sua contribui\u00e7\u00e3o. A camada de regras ajuda a organizar a decis\u00e3o, mas sua exist\u00eancia n\u00e3o resolve automaticamente a opacidade do modelo.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Um exemplo: quando a revis\u00e3o corrige a recomenda\u00e7\u00e3o<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Considere um caso hipot\u00e9tico de an\u00e1lise documental de sinistro.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A ap\u00f3lice tem in\u00edcio de vig\u00eancia em 15 de agosto. Um sistema de leitura interpreta a data do evento como 2 de agosto, embora o documento original registre 30 de agosto. A diverg\u00eancia aciona uma regra de encaminhamento para an\u00e1lise.<\/p>\n\n\n\n<style>\n  .tu-ai-decision-trace-table,\n  .tu-ai-decision-trace-table * {\n    box-sizing: border-box;\n  }\n\n  .tu-ai-decision-trace-table {\n    --tu-red: #DD4E4B;\n    --tu-black: #000000;\n    --tu-white: #FFFFFF;\n    --tu-soft: #FFF7F7;\n    --tu-line: #E7E7E7;\n    --tu-text: #333333;\n\n    width: 100%;\n    max-width: 100%;\n    margin: 24px 0;\n\n    font-family: \"DM Sans\", Arial, sans-serif;\n  }\n\n  .tu-ai-decision-trace-table__container {\n    width: 100%;\n    max-width: 100%;\n    overflow: hidden;\n\n    background: var(--tu-white);\n    border: 1px solid #D9D9D9;\n    border-radius: 4px 16px 16px 16px;\n  }\n\n  .tu-ai-decision-trace-table table {\n    width: 100% !important;\n    max-width: 100% !important;\n    margin: 0 !important;\n\n    border-collapse: separate;\n    border-spacing: 0;\n    table-layout: fixed;\n  }\n\n  .tu-ai-decision-trace-table th,\n  .tu-ai-decision-trace-table td {\n    padding: 14px 15px;\n\n    vertical-align: top;\n    text-align: left;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-decision-trace-table thead th {\n    background: var(--tu-red);\n    color: var(--tu-black);\n\n    border-right: 1px solid rgba(0, 0, 0, .15);\n\n    font-size: 11.9px;\n    font-weight: 700;\n    line-height: 1.4;\n  }\n\n  .tu-ai-decision-trace-table thead th:last-child {\n    border-right: 0;\n  }\n\n  .tu-ai-decision-trace-table thead th:nth-child(1) {\n    width: 23%;\n  }\n\n  .tu-ai-decision-trace-table thead th:nth-child(2) {\n    width: 34%;\n  }\n\n  .tu-ai-decision-trace-table thead th:nth-child(3) {\n    width: 43%;\n  }\n\n  .tu-ai-decision-trace-table tbody td {\n    position: relative;\n\n    background: var(--tu-white);\n\n    border-right: 1px solid var(--tu-line);\n    border-bottom: 1px solid var(--tu-line);\n\n    color: var(--tu-text);\n\n    font-size: 12.3px;\n    font-weight: 400;\n    line-height: 1.55;\n  }\n\n  .tu-ai-decision-trace-table tbody tr:nth-child(even) td {\n    background: var(--tu-soft);\n  }\n\n  .tu-ai-decision-trace-table tbody td:last-child {\n    border-right: 0;\n  }\n\n  .tu-ai-decision-trace-table tbody tr:last-child td {\n    border-bottom: 0;\n  }\n\n  .tu-ai-decision-trace-table tbody td:first-child {\n    padding-left: 19px;\n\n    color: var(--tu-black);\n    font-weight: 700;\n  }\n\n  .tu-ai-decision-trace-table tbody td:first-child::before {\n    content: \"\";\n\n    position: absolute;\n    top: 0;\n    bottom: 0;\n    left: 0;\n\n    width: 4px;\n\n    background: var(--tu-red);\n  }\n\n  .tu-ai-decision-trace-table__evidence {\n    color: var(--tu-black) !important;\n    font-weight: 600 !important;\n  }\n\n  @media (max-width: 680px) {\n    .tu-ai-decision-trace-table {\n      margin: 20px 0;\n    }\n\n    .tu-ai-decision-trace-table__container {\n      overflow: visible;\n\n      background: transparent;\n      border: 0;\n      border-radius: 0;\n    }\n\n    .tu-ai-decision-trace-table table,\n    .tu-ai-decision-trace-table tbody,\n    .tu-ai-decision-trace-table tr,\n    .tu-ai-decision-trace-table td {\n      display: block;\n\n      width: 100% !important;\n      max-width: 100% !important;\n    }\n\n    .tu-ai-decision-trace-table thead {\n      position: absolute;\n\n      width: 1px;\n      height: 1px;\n\n      margin: -1px;\n      padding: 0;\n\n      overflow: hidden;\n\n      clip: rect(0, 0, 0, 0);\n      white-space: nowrap;\n\n      border: 0;\n    }\n\n    .tu-ai-decision-trace-table tbody {\n      display: grid;\n      gap: 12px;\n    }\n\n    .tu-ai-decision-trace-table tbody tr {\n      overflow: hidden;\n\n      background: var(--tu-white);\n      border: 1px solid #DCDCDC;\n      border-radius: 4px 13px 13px 13px;\n    }\n\n    .tu-ai-decision-trace-table tbody td,\n    .tu-ai-decision-trace-table tbody tr:nth-child(even) td {\n      padding: 11px 14px;\n\n      background: var(--tu-white);\n\n      border-right: 0;\n      border-bottom: 1px solid var(--tu-line);\n\n      font-size: 12.7px;\n      line-height: 1.5;\n    }\n\n    .tu-ai-decision-trace-table tbody td:last-child {\n      border-bottom: 0;\n    }\n\n    .tu-ai-decision-trace-table tbody td::before {\n      display: block;\n\n      margin-bottom: 4px;\n\n      color: #6A6A6A;\n\n      font-size: 9.8px;\n      font-weight: 700;\n      line-height: 1.2;\n      letter-spacing: .035em;\n      text-transform: uppercase;\n\n      content: attr(data-label);\n    }\n\n    .tu-ai-decision-trace-table tbody td:first-child {\n      padding: 13px 14px;\n\n      background: var(--tu-red) !important;\n      color: var(--tu-black) !important;\n\n      font-size: 13.3px;\n      font-weight: 700;\n    }\n\n    .tu-ai-decision-trace-table tbody td:first-child::before {\n      position: static;\n\n      display: block;\n\n      width: auto;\n      height: auto;\n      margin-bottom: 4px;\n\n      background: none;\n\n      color: rgba(0, 0, 0, .65);\n\n      content: attr(data-label);\n    }\n  }\n<\/style>\n\n<div class=\"tu-ai-decision-trace-table\">\n\n  <div class=\"tu-ai-decision-trace-table__container\">\n\n    <table>\n\n      <thead>\n        <tr>\n          <th scope=\"col\">Etapa<\/th>\n          <th scope=\"col\">O que acontece<\/th>\n          <th scope=\"col\">O que precisa ficar dispon\u00edvel<\/th>\n        <\/tr>\n      <\/thead>\n\n      <tbody>\n\n        <tr>\n          <td data-label=\"Etapa\">\n            Documento recebido\n          <\/td>\n\n          <td data-label=\"O que acontece\">\n            O comprovante informa a data do evento.\n          <\/td>\n\n          <td\n            class=\"tu-ai-decision-trace-table__evidence\"\n            data-label=\"O que precisa ficar dispon\u00edvel\"\n          >\n            Arquivo original e identifica\u00e7\u00e3o de sua origem.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Etapa\">\n            Leitura automatizada\n          <\/td>\n\n          <td data-label=\"O que acontece\">\n            O sistema extrai uma data incorreta.\n          <\/td>\n\n          <td\n            class=\"tu-ai-decision-trace-table__evidence\"\n            data-label=\"O que precisa ficar dispon\u00edvel\"\n          >\n            Resultado extra\u00eddo e identifica\u00e7\u00e3o da vers\u00e3o utilizada.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Etapa\">\n            Regra aplicada\n          <\/td>\n\n          <td data-label=\"O que acontece\">\n            A data anterior \u00e0 vig\u00eancia gera uma pend\u00eancia.\n          <\/td>\n\n          <td\n            class=\"tu-ai-decision-trace-table__evidence\"\n            data-label=\"O que precisa ficar dispon\u00edvel\"\n          >\n            Regra e configura\u00e7\u00e3o vigentes.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Etapa\">\n            Revis\u00e3o humana\n          <\/td>\n\n          <td data-label=\"O que acontece\">\n            O analista consulta o original e corrige a extra\u00e7\u00e3o.\n          <\/td>\n\n          <td\n            class=\"tu-ai-decision-trace-table__evidence\"\n            data-label=\"O que precisa ficar dispon\u00edvel\"\n          >\n            Conte\u00fado apresentado, corre\u00e7\u00e3o e fundamento.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Etapa\">\n            Decis\u00e3o final\n          <\/td>\n\n          <td data-label=\"O que acontece\">\n            A cobertura \u00e9 confirmada ap\u00f3s as demais verifica\u00e7\u00f5es.\n          <\/td>\n\n          <td\n            class=\"tu-ai-decision-trace-table__evidence\"\n            data-label=\"O que precisa ficar dispon\u00edvel\"\n          >\n            Evid\u00eancias e fundamentos efetivamente utilizados.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Etapa\">\n            Comunica\u00e7\u00e3o\n          <\/td>\n\n          <td data-label=\"O que acontece\">\n            O segurado recebe a manifesta\u00e7\u00e3o correspondente.\n          <\/td>\n\n          <td\n            class=\"tu-ai-decision-trace-table__evidence\"\n            data-label=\"O que precisa ficar dispon\u00edvel\"\n          >\n            Conte\u00fado, data e registro do envio.\n          <\/td>\n        <\/tr>\n\n      <\/tbody>\n\n    <\/table>\n\n  <\/div>\n\n<\/div>\n\n\n\n<p class=\"wp-block-paragraph\">O exemplo mostra por que a tela do revisor precisa oferecer acesso \u00e0 evid\u00eancia original. Se ela exibir apenas uma recomenda\u00e7\u00e3o, a interven\u00e7\u00e3o humana pode n\u00e3o identificar a falha.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Tamb\u00e9m mostra por que corre\u00e7\u00f5es devem preservar o hist\u00f3rico. Sobrescrever a data extra\u00edda sem registrar a altera\u00e7\u00e3o apagaria a informa\u00e7\u00e3o necess\u00e1ria para entender o encaminhamento inicial.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">O que registrar para reconstituir uma decis\u00e3o<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Os registros precisam ser recuper\u00e1veis pelo identificador da proposta, ap\u00f3lice ou sinistro. Isso pode ser feito por v\u00ednculos entre sistemas, sem concentrar todos os dados em um \u00fanico reposit\u00f3rio.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">O conjunto deve permitir recuperar:<\/p>\n\n\n\n<style>\n  .tu-ai-traceability-records-list,\n  .tu-ai-traceability-records-list * {\n    box-sizing: border-box;\n  }\n\n  .tu-ai-traceability-records-list {\n    --tu-red: #DD4E4B;\n    --tu-black: #000000;\n    --tu-white: #FFFFFF;\n    --tu-soft: #FFF7F7;\n    --tu-line: #E7E7E7;\n    --tu-text: #383838;\n\n    width: 100%;\n    max-width: 100%;\n    margin: 24px 0;\n\n    font-family: \"DM Sans\", Arial, sans-serif;\n  }\n\n  .tu-ai-traceability-records-list__items {\n    position: relative;\n\n    display: flex;\n    flex-direction: column;\n    gap: 9px;\n\n    width: 100%;\n  }\n\n  .tu-ai-traceability-records-list__items::before {\n    content: \"\";\n\n    position: absolute;\n    top: 21px;\n    bottom: 21px;\n    left: 19px;\n\n    width: 2px;\n\n    background: #F0C1BF;\n  }\n\n  .tu-ai-traceability-records-list__item {\n    position: relative;\n    z-index: 1;\n\n    display: grid;\n    grid-template-columns: 40px minmax(0, 190px) minmax(0, 1fr);\n    gap: 13px;\n    align-items: start;\n\n    width: 100%;\n    min-width: 0;\n    padding: 14px 16px 14px 0;\n\n    background: var(--tu-white);\n    border: 1px solid var(--tu-line);\n    border-radius: 4px 13px 13px 13px;\n  }\n\n  .tu-ai-traceability-records-list__item:nth-child(even) {\n    background: var(--tu-soft);\n  }\n\n  .tu-ai-traceability-records-list__number {\n    display: flex;\n    align-items: center;\n    justify-content: center;\n\n    width: 38px;\n    height: 38px;\n\n    background: var(--tu-red);\n    border: 4px solid var(--tu-white);\n    border-radius: 50%;\n\n    color: var(--tu-white);\n    font-size: 9.5px;\n    font-weight: 700;\n    line-height: 1;\n  }\n\n  .tu-ai-traceability-records-list__item:nth-child(even)\n  .tu-ai-traceability-records-list__number {\n    border-color: var(--tu-soft);\n  }\n\n  .tu-ai-traceability-records-list__title {\n    margin: 8px 0 0 !important;\n    padding: 0 !important;\n\n    color: var(--tu-black);\n    font-size: 13.8px;\n    font-weight: 700;\n    line-height: 1.42;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-traceability-records-list__text {\n    margin: 7px 0 0 !important;\n    padding: 0 !important;\n\n    color: var(--tu-text) !important;\n    font-size: 12.8px;\n    font-weight: 400;\n    line-height: 1.6;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  @media (max-width: 820px) {\n    .tu-ai-traceability-records-list__item {\n      grid-template-columns: 40px minmax(0, 165px) minmax(0, 1fr);\n    }\n  }\n\n  @media (max-width: 680px) {\n    .tu-ai-traceability-records-list {\n      margin: 20px 0;\n    }\n\n    .tu-ai-traceability-records-list__items {\n      gap: 8px;\n    }\n\n    .tu-ai-traceability-records-list__items::before {\n      left: 17px;\n    }\n\n    .tu-ai-traceability-records-list__item {\n      grid-template-columns: 36px minmax(0, 1fr);\n      gap: 4px 10px;\n\n      padding: 13px 14px 13px 0;\n    }\n\n    .tu-ai-traceability-records-list__number {\n      width: 34px;\n      height: 34px;\n\n      font-size: 9px;\n    }\n\n    .tu-ai-traceability-records-list__title {\n      margin-top: 6px !important;\n      font-size: 13.4px;\n    }\n\n    .tu-ai-traceability-records-list__text {\n      grid-column: 2;\n\n      margin-top: 0 !important;\n\n      font-size: 12.5px;\n    }\n  }\n<\/style>\n\n<div class=\"tu-ai-traceability-records-list\">\n\n  <div class=\"tu-ai-traceability-records-list__items\">\n\n    <section class=\"tu-ai-traceability-records-list__item\">\n\n      <span class=\"tu-ai-traceability-records-list__number\" aria-hidden=\"true\">\n        01\n      <\/span>\n\n      <p class=\"tu-ai-traceability-records-list__title\">\n        Entradas utilizadas\n      <\/p>\n\n      <p class=\"tu-ai-traceability-records-list__text\">\n        Valores efetivamente considerados, fontes e momento da consulta, na medida necess\u00e1ria para o caso.\n      <\/p>\n\n    <\/section>\n\n\n    <section class=\"tu-ai-traceability-records-list__item\">\n\n      <span class=\"tu-ai-traceability-records-list__number\" aria-hidden=\"true\">\n        02\n      <\/span>\n\n      <p class=\"tu-ai-traceability-records-list__title\">\n        Prepara\u00e7\u00e3o dos dados\n      <\/p>\n\n      <p class=\"tu-ai-traceability-records-list__text\">\n        Transforma\u00e7\u00f5es, preenchimentos e valida\u00e7\u00f5es que alteraram o conte\u00fado recebido pelo modelo.\n      <\/p>\n\n    <\/section>\n\n\n    <section class=\"tu-ai-traceability-records-list__item\">\n\n      <span class=\"tu-ai-traceability-records-list__number\" aria-hidden=\"true\">\n        03\n      <\/span>\n\n      <p class=\"tu-ai-traceability-records-list__title\">\n        Vers\u00f5es e configura\u00e7\u00f5es\n      <\/p>\n\n      <p class=\"tu-ai-traceability-records-list__text\">\n        Modelo, regras, par\u00e2metros de corte e, quando aplic\u00e1vel, prompts e documentos de contexto.\n      <\/p>\n\n    <\/section>\n\n\n    <section class=\"tu-ai-traceability-records-list__item\">\n\n      <span class=\"tu-ai-traceability-records-list__number\" aria-hidden=\"true\">\n        04\n      <\/span>\n\n      <p class=\"tu-ai-traceability-records-list__title\">\n        Sa\u00edda original\n      <\/p>\n\n      <p class=\"tu-ai-traceability-records-list__text\">\n        Score, classe, extra\u00e7\u00e3o ou texto antes de sua apresenta\u00e7\u00e3o na interface.\n      <\/p>\n\n    <\/section>\n\n\n    <section class=\"tu-ai-traceability-records-list__item\">\n\n      <span class=\"tu-ai-traceability-records-list__number\" aria-hidden=\"true\">\n        05\n      <\/span>\n\n      <p class=\"tu-ai-traceability-records-list__title\">\n        Explica\u00e7\u00e3o e m\u00e9todo\n      <\/p>\n\n      <p class=\"tu-ai-traceability-records-list__text\">\n        Resultado apresentado ou recursos necess\u00e1rios para recalcul\u00e1-lo, incluindo configura\u00e7\u00e3o e conjunto de refer\u00eancia.\n      <\/p>\n\n    <\/section>\n\n\n    <section class=\"tu-ai-traceability-records-list__item\">\n\n      <span class=\"tu-ai-traceability-records-list__number\" aria-hidden=\"true\">\n        06\n      <\/span>\n\n      <p class=\"tu-ai-traceability-records-list__title\">\n        Interven\u00e7\u00e3o humana\n      <\/p>\n\n      <p class=\"tu-ai-traceability-records-list__text\">\n        Informa\u00e7\u00e3o dispon\u00edvel ao revisor, decis\u00e3o tomada e fundamento das altera\u00e7\u00f5es.\n      <\/p>\n\n    <\/section>\n\n\n    <section class=\"tu-ai-traceability-records-list__item\">\n\n      <span class=\"tu-ai-traceability-records-list__number\" aria-hidden=\"true\">\n        07\n      <\/span>\n\n      <p class=\"tu-ai-traceability-records-list__title\">\n        Comunica\u00e7\u00e3o\n      <\/p>\n\n      <p class=\"tu-ai-traceability-records-list__text\">\n        Conte\u00fado enviado, destinat\u00e1rio, data, canal e comprova\u00e7\u00e3o pertinente.\n      <\/p>\n\n    <\/section>\n\n  <\/div>\n\n<\/div>\n\n\n\n<p class=\"wp-block-paragraph\">Esses registros devem ser protegidos contra altera\u00e7\u00e3o indevida, com controle de acesso e hist\u00f3rico das corre\u00e7\u00f5es.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A capacidade de acompanhar um caso entre integra\u00e7\u00f5es, modelos e sistemas de neg\u00f3cio se relaciona \u00e0 <strong><a href=\"https:\/\/tecnologiaunica.com.br\/blog\/observabilidade-em-sistemas-de-seguros\/\" target=\"_blank\" rel=\"noreferrer noopener\">observabilidade em sistemas de seguros<\/a><\/strong>. Para explicabilidade, \u00e9 necess\u00e1rio conectar esse acompanhamento aos fundamentos da decis\u00e3o.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Persistir a explica\u00e7\u00e3o ou recalcular depois?<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">As duas op\u00e7\u00f5es podem ser adequadas, desde que suas condi\u00e7\u00f5es sejam verificadas.<\/p>\n\n\n\n<style>\n  .tu-ai-explanation-storage-strategies,\n  .tu-ai-explanation-storage-strategies * {\n    box-sizing: border-box;\n  }\n\n  .tu-ai-explanation-storage-strategies {\n    --tu-red: #DD4E4B;\n    --tu-black: #000000;\n    --tu-white: #FFFFFF;\n    --tu-soft: #FFF7F7;\n    --tu-line: #E7E7E7;\n    --tu-text: #333333;\n\n    width: 100%;\n    max-width: 100%;\n    margin: 24px 0;\n\n    font-family: \"DM Sans\", Arial, sans-serif;\n  }\n\n  .tu-ai-explanation-storage-strategies__container {\n    width: 100%;\n    max-width: 100%;\n    overflow: hidden;\n\n    background: var(--tu-white);\n    border: 1px solid #D9D9D9;\n    border-radius: 4px 16px 16px 16px;\n  }\n\n  .tu-ai-explanation-storage-strategies table {\n    width: 100% !important;\n    max-width: 100% !important;\n    margin: 0 !important;\n\n    border-collapse: separate;\n    border-spacing: 0;\n    table-layout: fixed;\n  }\n\n  .tu-ai-explanation-storage-strategies th,\n  .tu-ai-explanation-storage-strategies td {\n    padding: 14px 15px;\n\n    vertical-align: top;\n    text-align: left;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-explanation-storage-strategies thead th {\n    background: var(--tu-red);\n    color: var(--tu-black);\n\n    border-right: 1px solid rgba(0, 0, 0, .15);\n\n    font-size: 11.9px;\n    font-weight: 700;\n    line-height: 1.4;\n  }\n\n  .tu-ai-explanation-storage-strategies thead th:last-child {\n    border-right: 0;\n  }\n\n  .tu-ai-explanation-storage-strategies thead th:nth-child(1) {\n    width: 31%;\n  }\n\n  .tu-ai-explanation-storage-strategies thead th:nth-child(2) {\n    width: 34%;\n  }\n\n  .tu-ai-explanation-storage-strategies thead th:nth-child(3) {\n    width: 35%;\n  }\n\n  .tu-ai-explanation-storage-strategies tbody td {\n    position: relative;\n\n    background: var(--tu-white);\n\n    border-right: 1px solid var(--tu-line);\n    border-bottom: 1px solid var(--tu-line);\n\n    color: var(--tu-text);\n\n    font-size: 12.3px;\n    font-weight: 400;\n    line-height: 1.55;\n  }\n\n  .tu-ai-explanation-storage-strategies tbody tr:nth-child(even) td {\n    background: var(--tu-soft);\n  }\n\n  .tu-ai-explanation-storage-strategies tbody td:last-child {\n    border-right: 0;\n  }\n\n  .tu-ai-explanation-storage-strategies tbody tr:last-child td {\n    border-bottom: 0;\n  }\n\n  .tu-ai-explanation-storage-strategies tbody td:first-child {\n    padding-left: 19px;\n\n    color: var(--tu-black);\n    font-weight: 700;\n  }\n\n  .tu-ai-explanation-storage-strategies tbody td:first-child::before {\n    content: \"\";\n\n    position: absolute;\n    top: 0;\n    bottom: 0;\n    left: 0;\n\n    width: 4px;\n\n    background: var(--tu-red);\n  }\n\n  .tu-ai-explanation-storage-strategies__condition {\n    color: var(--tu-black) !important;\n    font-weight: 600 !important;\n  }\n\n  @media (max-width: 680px) {\n    .tu-ai-explanation-storage-strategies {\n      margin: 20px 0;\n    }\n\n    .tu-ai-explanation-storage-strategies__container {\n      overflow: visible;\n\n      background: transparent;\n      border: 0;\n      border-radius: 0;\n    }\n\n    .tu-ai-explanation-storage-strategies table,\n    .tu-ai-explanation-storage-strategies tbody,\n    .tu-ai-explanation-storage-strategies tr,\n    .tu-ai-explanation-storage-strategies td {\n      display: block;\n\n      width: 100% !important;\n      max-width: 100% !important;\n    }\n\n    .tu-ai-explanation-storage-strategies thead {\n      position: absolute;\n\n      width: 1px;\n      height: 1px;\n\n      margin: -1px;\n      padding: 0;\n\n      overflow: hidden;\n\n      clip: rect(0, 0, 0, 0);\n      white-space: nowrap;\n\n      border: 0;\n    }\n\n    .tu-ai-explanation-storage-strategies tbody {\n      display: grid;\n      gap: 12px;\n    }\n\n    .tu-ai-explanation-storage-strategies tbody tr {\n      overflow: hidden;\n\n      background: var(--tu-white);\n      border: 1px solid #DCDCDC;\n      border-radius: 4px 13px 13px 13px;\n    }\n\n    .tu-ai-explanation-storage-strategies tbody td,\n    .tu-ai-explanation-storage-strategies tbody tr:nth-child(even) td {\n      padding: 11px 14px;\n\n      background: var(--tu-white);\n\n      border-right: 0;\n      border-bottom: 1px solid var(--tu-line);\n\n      font-size: 12.7px;\n      line-height: 1.5;\n    }\n\n    .tu-ai-explanation-storage-strategies tbody td:last-child {\n      border-bottom: 0;\n    }\n\n    .tu-ai-explanation-storage-strategies tbody td::before {\n      display: block;\n\n      margin-bottom: 4px;\n\n      color: #6A6A6A;\n\n      font-size: 9.8px;\n      font-weight: 700;\n      line-height: 1.2;\n      letter-spacing: .035em;\n      text-transform: uppercase;\n\n      content: attr(data-label);\n    }\n\n    .tu-ai-explanation-storage-strategies tbody td:first-child {\n      padding: 13px 14px;\n\n      background: var(--tu-red) !important;\n      color: var(--tu-black) !important;\n\n      font-size: 13.3px;\n      font-weight: 700;\n    }\n\n    .tu-ai-explanation-storage-strategies tbody td:first-child::before {\n      position: static;\n\n      display: block;\n\n      width: auto;\n      height: auto;\n      margin-bottom: 4px;\n\n      background: none;\n\n      color: rgba(0, 0, 0, .65);\n\n      content: attr(data-label);\n    }\n  }\n<\/style>\n\n<div class=\"tu-ai-explanation-storage-strategies\">\n\n  <div class=\"tu-ai-explanation-storage-strategies__container\">\n\n    <table>\n\n      <thead>\n        <tr>\n          <th scope=\"col\">Estrat\u00e9gia<\/th>\n          <th scope=\"col\">Quando avaliar<\/th>\n          <th scope=\"col\">Condi\u00e7\u00e3o essencial<\/th>\n        <\/tr>\n      <\/thead>\n\n      <tbody>\n\n        <tr>\n          <td data-label=\"Estrat\u00e9gia\">\n            Persistir a explica\u00e7\u00e3o junto da decis\u00e3o\n          <\/td>\n\n          <td data-label=\"Quando avaliar\">\n            Necessidade de resposta r\u00e1pida ou dificuldade de reproduzir o ambiente.\n          <\/td>\n\n          <td\n            class=\"tu-ai-explanation-storage-strategies__condition\"\n            data-label=\"Condi\u00e7\u00e3o essencial\"\n          >\n            Validar a explica\u00e7\u00e3o armazenada e preservar sua configura\u00e7\u00e3o.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Estrat\u00e9gia\">\n            Recalcular quando necess\u00e1rio\n          <\/td>\n\n          <td data-label=\"Quando avaliar\">\n            Demanda eventual e custo elevado de c\u00e1lculo cont\u00ednuo.\n          <\/td>\n\n          <td\n            class=\"tu-ai-explanation-storage-strategies__condition\"\n            data-label=\"Condi\u00e7\u00e3o essencial\"\n          >\n            Preservar e testar o carregamento do modelo, dados e m\u00e9todo originais.\n          <\/td>\n        <\/tr>\n\n        <tr>\n          <td data-label=\"Estrat\u00e9gia\">\n            Combinar estrat\u00e9gias\n          <\/td>\n\n          <td data-label=\"Quando avaliar\">\n            Decis\u00f5es com efeitos e volumes diferentes.\n          <\/td>\n\n          <td\n            class=\"tu-ai-explanation-storage-strategies__condition\"\n            data-label=\"Condi\u00e7\u00e3o essencial\"\n          >\n            Definir crit\u00e9rios documentados para cada grupo.\n          <\/td>\n        <\/tr>\n\n      <\/tbody>\n\n    <\/table>\n\n  <\/div>\n\n<\/div>\n\n\n\n<p class=\"wp-block-paragraph\">Recalcular com o modelo atual n\u00e3o explica automaticamente uma execu\u00e7\u00e3o antiga. Mesmo com o modelo preservado, mudan\u00e7as no m\u00e9todo de explica\u00e7\u00e3o ou em seus dados de refer\u00eancia podem alterar o resultado.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Em sistemas generativos, reconstituir o que ocorreu tamb\u00e9m exige preservar a sa\u00edda efetivamente utilizada. Uma nova gera\u00e7\u00e3o pode produzir outro texto.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Como verificar se a explica\u00e7\u00e3o funciona<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">A exist\u00eancia de um arquivo de explica\u00e7\u00e3o \u00e9 apenas o come\u00e7o. Sua avalia\u00e7\u00e3o deve responder a quatro perguntas:<\/p>\n\n\n\n<style>\n  .tu-ai-evaluation-criteria,\n  .tu-ai-evaluation-criteria * {\n    box-sizing: border-box;\n  }\n\n  .tu-ai-evaluation-criteria {\n    --tu-red: #DD4E4B;\n    --tu-black: #000000;\n    --tu-white: #FFFFFF;\n    --tu-soft: #FFF7F7;\n    --tu-line: #E7E7E7;\n    --tu-text: #383838;\n\n    width: 100%;\n    max-width: 100%;\n    margin: 24px 0;\n\n    font-family: \"DM Sans\", Arial, sans-serif;\n  }\n\n  .tu-ai-evaluation-criteria__grid {\n    display: grid;\n    grid-template-columns: repeat(2, minmax(0, 1fr));\n    gap: 12px;\n\n    width: 100%;\n  }\n\n  .tu-ai-evaluation-criteria__item {\n    display: grid;\n    grid-template-columns: 34px minmax(0, 1fr);\n    gap: 11px;\n    align-items: start;\n\n    min-width: 0;\n    padding: 17px 16px;\n\n    background: var(--tu-white);\n    border: 1px solid var(--tu-line);\n    border-radius: 4px 14px 14px 14px;\n  }\n\n  .tu-ai-evaluation-criteria__item:nth-child(even) {\n    background: var(--tu-soft);\n  }\n\n  .tu-ai-evaluation-criteria__marker {\n    display: flex;\n    align-items: center;\n    justify-content: center;\n\n    width: 30px;\n    height: 30px;\n\n    background: var(--tu-red);\n    border-radius: 50%;\n\n    color: var(--tu-white);\n    font-size: 13px;\n    font-weight: 700;\n    line-height: 1;\n  }\n\n  .tu-ai-evaluation-criteria__content {\n    min-width: 0;\n  }\n\n  .tu-ai-evaluation-criteria__title {\n    margin: 1px 0 5px !important;\n    padding: 0 !important;\n\n    color: var(--tu-black);\n    font-size: 13.8px;\n    font-weight: 700;\n    line-height: 1.4;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-evaluation-criteria__text {\n    margin: 0 !important;\n    padding: 0 !important;\n\n    color: var(--tu-text) !important;\n    font-size: 12.75px;\n    font-weight: 400;\n    line-height: 1.58;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  @media (max-width: 720px) {\n    .tu-ai-evaluation-criteria__grid {\n      grid-template-columns: 1fr;\n      gap: 9px;\n    }\n\n    .tu-ai-evaluation-criteria {\n      margin: 20px 0;\n    }\n\n    .tu-ai-evaluation-criteria__item {\n      padding: 15px 14px;\n    }\n\n    .tu-ai-evaluation-criteria__marker {\n      width: 28px;\n      height: 28px;\n\n      font-size: 12px;\n    }\n\n    .tu-ai-evaluation-criteria__title {\n      font-size: 13.4px;\n    }\n\n    .tu-ai-evaluation-criteria__text {\n      font-size: 12.5px;\n    }\n  }\n<\/style>\n\n<div class=\"tu-ai-evaluation-criteria\">\n\n  <div class=\"tu-ai-evaluation-criteria__grid\">\n\n    <section class=\"tu-ai-evaluation-criteria__item\">\n\n      <span class=\"tu-ai-evaluation-criteria__marker\" aria-hidden=\"true\">\n        \u2713\n      <\/span>\n\n      <div class=\"tu-ai-evaluation-criteria__content\">\n        <p class=\"tu-ai-evaluation-criteria__title\">\n          Fidelidade\n        <\/p>\n\n        <p class=\"tu-ai-evaluation-criteria__text\">\n          O conte\u00fado corresponde ao modelo, \u00e0s regras e \u00e0s evid\u00eancias daquele caso?\n        <\/p>\n      <\/div>\n\n    <\/section>\n\n\n    <section class=\"tu-ai-evaluation-criteria__item\">\n\n      <span class=\"tu-ai-evaluation-criteria__marker\" aria-hidden=\"true\">\n        \u2713\n      <\/span>\n\n      <div class=\"tu-ai-evaluation-criteria__content\">\n        <p class=\"tu-ai-evaluation-criteria__title\">\n          Compreens\u00e3o\n        <\/p>\n\n        <p class=\"tu-ai-evaluation-criteria__text\">\n          O destinat\u00e1rio consegue identificar o motivo apresentado?\n        <\/p>\n      <\/div>\n\n    <\/section>\n\n\n    <section class=\"tu-ai-evaluation-criteria__item\">\n\n      <span class=\"tu-ai-evaluation-criteria__marker\" aria-hidden=\"true\">\n        \u2713\n      <\/span>\n\n      <div class=\"tu-ai-evaluation-criteria__content\">\n        <p class=\"tu-ai-evaluation-criteria__title\">\n          Utilidade\n        <\/p>\n\n        <p class=\"tu-ai-evaluation-criteria__text\">\n          O revisor consegue detectar erros e fundamentar uma interven\u00e7\u00e3o?\n        <\/p>\n      <\/div>\n\n    <\/section>\n\n\n    <section class=\"tu-ai-evaluation-criteria__item\">\n\n      <span class=\"tu-ai-evaluation-criteria__marker\" aria-hidden=\"true\">\n        \u2713\n      <\/span>\n\n      <div class=\"tu-ai-evaluation-criteria__content\">\n        <p class=\"tu-ai-evaluation-criteria__title\">\n          Recupera\u00e7\u00e3o\n        <\/p>\n\n        <p class=\"tu-ai-evaluation-criteria__text\">\n          A opera\u00e7\u00e3o consegue reunir as evid\u00eancias dentro do prazo necess\u00e1rio?\n        <\/p>\n      <\/div>\n\n    <\/section>\n\n  <\/div>\n\n<\/div>\n\n\n\n<p class=\"wp-block-paragraph\">A taxa de diverg\u00eancia entre analista e modelo pode ajudar no acompanhamento, mas n\u00e3o prova a qualidade da revis\u00e3o. Poucas altera\u00e7\u00f5es podem refletir recomenda\u00e7\u00f5es adequadas ou dificuldade para perceber e corrigir falhas. \u00c9 necess\u00e1rio examinar casos e fundamentos.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">O n\u00edvel de detalhe tamb\u00e9m precisa considerar o destinat\u00e1rio. Em antifraude, divulgar limiares e mecanismos de detec\u00e7\u00e3o pode facilitar tentativas de contorno. A opera\u00e7\u00e3o deve definir quais informa\u00e7\u00f5es s\u00e3o necess\u00e1rias para esclarecer a decis\u00e3o e quais detalhes exigem prote\u00e7\u00e3o, com fundamento aplic\u00e1vel a cada situa\u00e7\u00e3o.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Checklist para avaliar uma decis\u00e3o real<\/h2>\n\n\n\n<style>\n  .tu-ai-traceability-checklist,\n  .tu-ai-traceability-checklist * {\n    box-sizing: border-box;\n  }\n\n  .tu-ai-traceability-checklist {\n    --tu-red: #DD4E4B;\n    --tu-coral: #E47F79;\n    --tu-black: #000000;\n    --tu-white: #FFFFFF;\n    --tu-soft: #FFF7F7;\n    --tu-line: #E7E7E7;\n    --tu-gray: #6A6A6A;\n\n    width: 100%;\n    max-width: 100%;\n    margin: 24px 0;\n\n    font-family: \"DM Sans\", Arial, sans-serif;\n  }\n\n  .tu-ai-traceability-checklist__box {\n    width: 100%;\n    padding: 24px;\n\n    background:\n      radial-gradient(circle at 100% 0%, rgba(221,78,75,.08), transparent 30%),\n      #FFFFFF;\n\n    border: 1px solid #D9D9D9;\n    border-radius: 4px 18px 18px 18px;\n  }\n\n  .tu-ai-traceability-checklist__intro {\n    display: flex;\n    align-items: flex-start;\n    justify-content: space-between;\n    gap: 20px;\n\n    margin-bottom: 18px;\n  }\n\n  .tu-ai-traceability-checklist__intro-text {\n    min-width: 0;\n  }\n\n  .tu-ai-traceability-checklist__eyebrow {\n    display: inline-flex;\n    align-items: center;\n    gap: 7px;\n\n    margin: 0 0 7px !important;\n    padding: 0 !important;\n\n    color: var(--tu-red) !important;\n\n    font-size: 10.4px;\n    font-weight: 700;\n    line-height: 1.2;\n    letter-spacing: .08em;\n    text-transform: uppercase;\n  }\n\n  .tu-ai-traceability-checklist__eyebrow::before {\n    content: \"\";\n\n    width: 18px;\n    height: 2px;\n\n    background: var(--tu-red);\n  }\n\n  .tu-ai-traceability-checklist__title {\n    margin: 0 !important;\n    padding: 0 !important;\n\n    color: var(--tu-black) !important;\n\n    font-size: 17px;\n    font-weight: 700;\n    line-height: 1.4;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-traceability-checklist__subtitle {\n    margin: 7px 0 0 !important;\n    padding: 0 !important;\n\n    max-width: 760px;\n\n    color: #484848 !important;\n\n    font-size: 12.9px;\n    font-weight: 400;\n    line-height: 1.6;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-traceability-checklist__score {\n    flex: 0 0 auto;\n\n    min-width: 92px;\n    padding: 10px 12px;\n\n    background: var(--tu-black);\n    border-radius: 4px 12px 12px 12px;\n\n    color: var(--tu-white);\n    text-align: center;\n  }\n\n  .tu-ai-traceability-checklist__score strong {\n    display: block;\n\n    font-size: 18px;\n    font-weight: 700;\n    line-height: 1;\n  }\n\n  .tu-ai-traceability-checklist__score span {\n    display: block;\n    margin-top: 4px;\n\n    color: #D3D3D3;\n\n    font-size: 9.5px;\n    font-weight: 600;\n    line-height: 1.2;\n    letter-spacing: .04em;\n    text-transform: uppercase;\n  }\n\n  .tu-ai-traceability-checklist__progress {\n    width: 100%;\n    height: 7px;\n    margin-bottom: 18px;\n\n    overflow: hidden;\n\n    background: #EEEEEE;\n    border-radius: 999px;\n  }\n\n  .tu-ai-traceability-checklist__progress-bar {\n    width: 0;\n    height: 100%;\n\n    background: var(--tu-red);\n    border-radius: inherit;\n\n    transition: width .25s ease;\n  }\n\n  .tu-ai-traceability-checklist__items {\n    display: grid;\n    grid-template-columns: repeat(2, minmax(0, 1fr));\n    gap: 10px;\n\n    width: 100%;\n  }\n\n  .tu-ai-traceability-checklist__item {\n    display: flex;\n    align-items: flex-start;\n    gap: 11px;\n\n    min-width: 0;\n    padding: 14px 14px;\n\n    background: var(--tu-white);\n    border: 1px solid var(--tu-line);\n    border-radius: 4px 12px 12px 12px;\n\n    cursor: pointer;\n\n    transition:\n      background-color .2s ease,\n      border-color .2s ease,\n      transform .2s ease;\n  }\n\n  .tu-ai-traceability-checklist__item:nth-child(4n+2),\n  .tu-ai-traceability-checklist__item:nth-child(4n+3) {\n    background: var(--tu-soft);\n  }\n\n  .tu-ai-traceability-checklist__item:hover {\n    border-color: #E6B2B0;\n  }\n\n  .tu-ai-traceability-checklist__item.is-checked {\n    background: #FFF1F0 !important;\n    border-color: var(--tu-red);\n  }\n\n  .tu-ai-traceability-checklist__input {\n    position: absolute;\n\n    width: 1px;\n    height: 1px;\n\n    opacity: 0;\n    pointer-events: none;\n  }\n\n  .tu-ai-traceability-checklist__check {\n    position: relative;\n\n    flex: 0 0 22px;\n    width: 22px;\n    height: 22px;\n    margin-top: 1px;\n\n    background: var(--tu-white);\n    border: 1.5px solid #B9B9B9;\n    border-radius: 6px;\n  }\n\n  .tu-ai-traceability-checklist__input:checked\n  + .tu-ai-traceability-checklist__check {\n    background: var(--tu-red);\n    border-color: var(--tu-red);\n  }\n\n  .tu-ai-traceability-checklist__input:checked\n  + .tu-ai-traceability-checklist__check::after {\n    content: \"\";\n\n    position: absolute;\n    top: 4px;\n    left: 7px;\n\n    width: 5px;\n    height: 9px;\n\n    border: solid #FFFFFF;\n    border-width: 0 2px 2px 0;\n\n    transform: rotate(45deg);\n  }\n\n  .tu-ai-traceability-checklist__input:focus-visible\n  + .tu-ai-traceability-checklist__check {\n    outline: 2px solid var(--tu-black);\n    outline-offset: 2px;\n  }\n\n  .tu-ai-traceability-checklist__question {\n    margin: 0 !important;\n    padding: 0 !important;\n\n    color: #303030 !important;\n\n    font-size: 12.6px;\n    font-weight: 500;\n    line-height: 1.55;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-traceability-checklist__footer {\n    display: grid;\n    grid-template-columns: minmax(0, 1fr) auto;\n    gap: 18px;\n    align-items: center;\n\n    margin-top: 18px;\n    padding-top: 17px;\n\n    border-top: 1px solid var(--tu-line);\n  }\n\n  .tu-ai-traceability-checklist__diagnosis {\n    min-width: 0;\n  }\n\n  .tu-ai-traceability-checklist__diagnosis strong {\n    display: block;\n    margin-bottom: 4px;\n\n    color: var(--tu-black);\n\n    font-size: 13.4px;\n    font-weight: 700;\n    line-height: 1.4;\n  }\n\n  .tu-ai-traceability-checklist__diagnosis p {\n    margin: 0 !important;\n    padding: 0 !important;\n\n    color: #555555 !important;\n\n    font-size: 12.4px;\n    font-weight: 400;\n    line-height: 1.55;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-traceability-checklist__reset {\n    appearance: none;\n\n    padding: 8px 11px;\n\n    background: transparent;\n    border: 1px solid #CACACA;\n    border-radius: 7px;\n\n    color: #555555;\n\n    font-family: inherit;\n    font-size: 11px;\n    font-weight: 600;\n    line-height: 1.2;\n\n    cursor: pointer;\n  }\n\n  .tu-ai-traceability-checklist__reset:hover {\n    border-color: var(--tu-black);\n    color: var(--tu-black);\n  }\n\n  .tu-ai-traceability-checklist__cta {\n    display: flex;\n    align-items: center;\n    justify-content: space-between;\n    gap: 22px;\n\n    width: 100%;\n    margin-top: 14px;\n    padding: 19px 20px;\n\n    background: var(--tu-black);\n    border: 1px solid var(--tu-red);\n    border-radius: 4px 14px 14px 14px;\n  }\n\n  .tu-ai-traceability-checklist__cta-content {\n    min-width: 0;\n  }\n\n  .tu-ai-traceability-checklist__cta-content strong {\n    display: block;\n    margin-bottom: 5px;\n\n    color: var(--tu-white);\n\n    font-size: 13.9px;\n    font-weight: 700;\n    line-height: 1.42;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-traceability-checklist__cta-content p {\n    margin: 0 !important;\n    padding: 0 !important;\n\n    max-width: 720px;\n\n    color: #D8D8D8 !important;\n\n    font-size: 12.6px;\n    font-weight: 400;\n    line-height: 1.58;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-traceability-checklist__button {\n    display: inline-flex;\n    align-items: center;\n    justify-content: center;\n    gap: 8px;\n\n    flex: 0 0 auto;\n\n    min-height: 42px;\n    padding: 10px 16px;\n\n    background: var(--tu-red);\n    border: 1px solid var(--tu-red);\n    border-radius: 8px;\n\n    color: var(--tu-white) !important;\n    text-decoration: none !important;\n\n    font-size: 12.4px;\n    font-weight: 700;\n    line-height: 1.2;\n\n    transition:\n      background-color .2s ease,\n      border-color .2s ease,\n      color .2s ease,\n      transform .2s ease;\n  }\n\n  .tu-ai-traceability-checklist__button::after {\n    content: \"\u2192\";\n    font-size: 15px;\n    line-height: 1;\n  }\n\n  .tu-ai-traceability-checklist__button:hover,\n  .tu-ai-traceability-checklist__button:focus {\n    background: #FFFFFF;\n    border-color: #FFFFFF;\n    color: #000000 !important;\n    text-decoration: none !important;\n  }\n\n  .tu-ai-traceability-checklist__button:focus-visible {\n    outline: 2px solid #FFFFFF;\n    outline-offset: 3px;\n  }\n\n  @media (hover: hover) and (pointer: fine) {\n    .tu-ai-traceability-checklist__button:hover {\n      transform: translateY(-1px);\n    }\n  }\n\n  @media (max-width: 760px) {\n    .tu-ai-traceability-checklist__items {\n      grid-template-columns: 1fr;\n    }\n\n    .tu-ai-traceability-checklist__item:nth-child(n) {\n      background: var(--tu-white);\n    }\n\n    .tu-ai-traceability-checklist__item:nth-child(even) {\n      background: var(--tu-soft);\n    }\n  }\n\n  @media (max-width: 620px) {\n    .tu-ai-traceability-checklist {\n      margin: 20px 0;\n    }\n\n    .tu-ai-traceability-checklist__box {\n      padding: 18px 16px;\n      border-radius: 4px 14px 14px 14px;\n    }\n\n    .tu-ai-traceability-checklist__intro {\n      flex-direction: column;\n      gap: 12px;\n    }\n\n    .tu-ai-traceability-checklist__score {\n      min-width: 84px;\n    }\n\n    .tu-ai-traceability-checklist__footer {\n      grid-template-columns: 1fr;\n      gap: 12px;\n    }\n\n    .tu-ai-traceability-checklist__reset {\n      width: 100%;\n    }\n\n    .tu-ai-traceability-checklist__cta {\n      flex-direction: column;\n      align-items: flex-start;\n      gap: 15px;\n\n      padding: 17px 16px;\n    }\n\n    .tu-ai-traceability-checklist__button {\n      width: 100%;\n    }\n  }\n<\/style>\n\n<div class=\"tu-ai-traceability-checklist\">\n\n  <div class=\"tu-ai-traceability-checklist__box\">\n\n    <div class=\"tu-ai-traceability-checklist__intro\">\n\n      <div class=\"tu-ai-traceability-checklist__intro-text\">\n\n        <p class=\"tu-ai-traceability-checklist__eyebrow\">\n          Diagn\u00f3stico r\u00e1pido\n        <\/p>\n\n        <p class=\"tu-ai-traceability-checklist__title\">\n          Escolha um caso conclu\u00eddo e verifique quantas respostas s\u00e3o \u201csim\u201d.\n        <\/p>\n\n        <p class=\"tu-ai-traceability-checklist__subtitle\">\n          O objetivo \u00e9 testar se a decis\u00e3o pode ser compreendida, reconstru\u00edda e sustentada a partir dos registros efetivamente dispon\u00edveis.\n        <\/p>\n\n      <\/div>\n\n      <div class=\"tu-ai-traceability-checklist__score\" aria-live=\"polite\">\n        <strong>\n          <span class=\"tu-ai-traceability-checklist__count\" style=\"display:inline;color:#FFFFFF;font-size:18px;letter-spacing:0;text-transform:none;\">0<\/span>\/9\n        <\/strong>\n        <span>crit\u00e9rios atendidos<\/span>\n      <\/div>\n\n    <\/div>\n\n\n    <div\n      class=\"tu-ai-traceability-checklist__progress\"\n      aria-hidden=\"true\"\n    >\n      <div class=\"tu-ai-traceability-checklist__progress-bar\"><\/div>\n    <\/div>\n\n\n    <div class=\"tu-ai-traceability-checklist__items\">\n\n      <label class=\"tu-ai-traceability-checklist__item\">\n        <input class=\"tu-ai-traceability-checklist__input\" type=\"checkbox\">\n        <span class=\"tu-ai-traceability-checklist__check\" aria-hidden=\"true\"><\/span>\n        <p class=\"tu-ai-traceability-checklist__question\">\n          \u00c9 poss\u00edvel identificar os dados e documentos efetivamente utilizados?\n        <\/p>\n      <\/label>\n\n      <label class=\"tu-ai-traceability-checklist__item\">\n        <input class=\"tu-ai-traceability-checklist__input\" type=\"checkbox\">\n        <span class=\"tu-ai-traceability-checklist__check\" aria-hidden=\"true\"><\/span>\n        <p class=\"tu-ai-traceability-checklist__question\">\n          As vers\u00f5es do modelo, das regras e das configura\u00e7\u00f5es est\u00e3o dispon\u00edveis?\n        <\/p>\n      <\/label>\n\n      <label class=\"tu-ai-traceability-checklist__item\">\n        <input class=\"tu-ai-traceability-checklist__input\" type=\"checkbox\">\n        <span class=\"tu-ai-traceability-checklist__check\" aria-hidden=\"true\"><\/span>\n        <p class=\"tu-ai-traceability-checklist__question\">\n          A sa\u00edda original foi preservada?\n        <\/p>\n      <\/label>\n\n      <label class=\"tu-ai-traceability-checklist__item\">\n        <input class=\"tu-ai-traceability-checklist__input\" type=\"checkbox\">\n        <span class=\"tu-ai-traceability-checklist__check\" aria-hidden=\"true\"><\/span>\n        <p class=\"tu-ai-traceability-checklist__question\">\n          A explica\u00e7\u00e3o apresentada pode ser recuperada ou recalculada com fidelidade?\n        <\/p>\n      <\/label>\n\n      <label class=\"tu-ai-traceability-checklist__item\">\n        <input class=\"tu-ai-traceability-checklist__input\" type=\"checkbox\">\n        <span class=\"tu-ai-traceability-checklist__check\" aria-hidden=\"true\"><\/span>\n        <p class=\"tu-ai-traceability-checklist__question\">\n          H\u00e1 v\u00ednculo entre os fatos do caso, o crit\u00e9rio aplicado e a decis\u00e3o final?\n        <\/p>\n      <\/label>\n\n      <label class=\"tu-ai-traceability-checklist__item\">\n        <input class=\"tu-ai-traceability-checklist__input\" type=\"checkbox\">\n        <span class=\"tu-ai-traceability-checklist__check\" aria-hidden=\"true\"><\/span>\n        <p class=\"tu-ai-traceability-checklist__question\">\n          Quando houve revis\u00e3o humana, est\u00e3o dispon\u00edveis as informa\u00e7\u00f5es vistas pelo revisor e o fundamento de sua atua\u00e7\u00e3o?\n        <\/p>\n      <\/label>\n\n      <label class=\"tu-ai-traceability-checklist__item\">\n        <input class=\"tu-ai-traceability-checklist__input\" type=\"checkbox\">\n        <span class=\"tu-ai-traceability-checklist__check\" aria-hidden=\"true\"><\/span>\n        <p class=\"tu-ai-traceability-checklist__question\">\n          A comunica\u00e7\u00e3o ao interessado pode ser recuperada?\n        <\/p>\n      <\/label>\n\n      <label class=\"tu-ai-traceability-checklist__item\">\n        <input class=\"tu-ai-traceability-checklist__input\" type=\"checkbox\">\n        <span class=\"tu-ai-traceability-checklist__check\" aria-hidden=\"true\"><\/span>\n        <p class=\"tu-ai-traceability-checklist__question\">\n          Os registros t\u00eam acesso controlado, reten\u00e7\u00e3o definida e prote\u00e7\u00e3o contra altera\u00e7\u00e3o indevida?\n        <\/p>\n      <\/label>\n\n      <label class=\"tu-ai-traceability-checklist__item\">\n        <input class=\"tu-ai-traceability-checklist__input\" type=\"checkbox\">\n        <span class=\"tu-ai-traceability-checklist__check\" aria-hidden=\"true\"><\/span>\n        <p class=\"tu-ai-traceability-checklist__question\">\n          A qualidade da explica\u00e7\u00e3o j\u00e1 foi verificada com os destinat\u00e1rios pertinentes?\n        <\/p>\n      <\/label>\n\n    <\/div>\n\n\n    <div class=\"tu-ai-traceability-checklist__footer\">\n\n      <div class=\"tu-ai-traceability-checklist__diagnosis\" aria-live=\"polite\">\n        <strong class=\"tu-ai-traceability-checklist__diagnosis-title\">\n          Comece pelo caso real.\n        <\/strong>\n\n        <p class=\"tu-ai-traceability-checklist__diagnosis-text\">\n          Marque apenas os crit\u00e9rios que podem ser comprovados com registros dispon\u00edveis hoje.\n        <\/p>\n      <\/div>\n\n      <button\n        class=\"tu-ai-traceability-checklist__reset\"\n        type=\"button\"\n      >\n        Limpar checklist\n      <\/button>\n\n    <\/div>\n\n  <\/div>\n\n\n  <div class=\"tu-ai-traceability-checklist__cta\">\n\n    <div class=\"tu-ai-traceability-checklist__cta-content\">\n\n      <strong>\n        Encontrou lacunas entre decis\u00e3o, explica\u00e7\u00e3o e evid\u00eancia?\n      <\/strong>\n\n      <p>\n        A Tecnologia \u00danica pode apoiar a estrutura\u00e7\u00e3o de sistemas, integra\u00e7\u00f5es e fluxos de dados para que informa\u00e7\u00f5es relevantes permane\u00e7am conectadas e recuper\u00e1veis ao longo da opera\u00e7\u00e3o.\n      <\/p>\n\n    <\/div>\n\n    <a\n      class=\"tu-ai-traceability-checklist__button\"\n      href=\"https:\/\/tecnologiaunica.com.br\/contato\/?utm_source=blog&amp;utm_medium=artigo&amp;utm_content=cta-checklist-rastreabilidade-ia\"\n      aria-label=\"Converse com a Tecnologia \u00danica sobre rastreabilidade, sistemas e integra\u00e7\u00f5es\"\n    >\n      Avaliar arquitetura e rastreabilidade\n    <\/a>\n\n  <\/div>\n\n<\/div>\n\n<script>\n(function () {\n  const root = document.currentScript.previousElementSibling;\n\n  if (!root || !root.classList.contains(\"tu-ai-traceability-checklist\")) return;\n\n  const inputs = root.querySelectorAll(\".tu-ai-traceability-checklist__input\");\n  const count = root.querySelector(\".tu-ai-traceability-checklist__count\");\n  const progress = root.querySelector(\".tu-ai-traceability-checklist__progress-bar\");\n  const diagnosisTitle = root.querySelector(\".tu-ai-traceability-checklist__diagnosis-title\");\n  const diagnosisText = root.querySelector(\".tu-ai-traceability-checklist__diagnosis-text\");\n  const reset = root.querySelector(\".tu-ai-traceability-checklist__reset\");\n\n  function updateChecklist() {\n    const checked = Array.from(inputs).filter(input => input.checked).length;\n    const total = inputs.length;\n    const percent = total ? (checked \/ total) * 100 : 0;\n\n    count.textContent = checked;\n    progress.style.width = percent + \"%\";\n\n    inputs.forEach(input => {\n      const item = input.closest(\".tu-ai-traceability-checklist__item\");\n      item.classList.toggle(\"is-checked\", input.checked);\n    });\n\n    if (checked === 0) {\n      diagnosisTitle.textContent = \"Comece pelo caso real.\";\n      diagnosisText.textContent = \"Marque apenas os crit\u00e9rios que podem ser comprovados com registros dispon\u00edveis hoje.\";\n    } else if (checked <= 3) {\n      diagnosisTitle.textContent = \"H\u00e1 lacunas relevantes de rastreabilidade.\";\n      diagnosisText.textContent = \"Priorize os registros que permitem reconstruir a entrada, a regra aplicada, a decis\u00e3o e eventuais interven\u00e7\u00f5es humanas.\";\n    } else if (checked <= 6) {\n      diagnosisTitle.textContent = \"A base existe, mas ainda h\u00e1 pontos fr\u00e1geis.\";\n      diagnosisText.textContent = \"Identifique quais etapas dependem de reconstru\u00e7\u00e3o manual ou n\u00e3o preservam vers\u00f5es, fundamentos e comunica\u00e7\u00f5es.\";\n    } else if (checked <= 8) {\n      diagnosisTitle.textContent = \"A rastreabilidade est\u00e1 avan\u00e7ada.\";\n      diagnosisText.textContent = \"Concentre a revis\u00e3o nos crit\u00e9rios ainda n\u00e3o comprov\u00e1veis e teste se as evid\u00eancias podem ser recuperadas no prazo necess\u00e1rio.\";\n    } else {\n      diagnosisTitle.textContent = \"Os nove crit\u00e9rios foram atendidos neste caso.\";\n      diagnosisText.textContent = \"Repita o teste em outros tipos de decis\u00e3o e verifique se o mesmo n\u00edvel de evid\u00eancia se mant\u00e9m de forma consistente.\";\n    }\n  }\n\n  inputs.forEach(input => {\n    input.addEventListener(\"change\", updateChecklist);\n  });\n\n  reset.addEventListener(\"click\", function () {\n    inputs.forEach(input => {\n      input.checked = false;\n    });\n\n    updateChecklist();\n  });\n\n  updateChecklist();\n})();\n<\/script>\n\n\n\n<p class=\"wp-block-paragraph\">Cada resposta negativa indica uma lacuna a investigar. A contagem n\u00e3o constitui nota de maturidade nem atestado de conformidade.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Os registros podem conter dados pessoais e sens\u00edveis. Por isso, o desenho deve contemplar finalidade, base legal, necessidade, seguran\u00e7a e descarte, conforme as disposi\u00e7\u00f5es aplic\u00e1veis da <a href=\"https:\/\/www.planalto.gov.br\/ccivil_03\/_ato2015-2018\/2018\/lei\/l13709.htm\">LGPD<\/a>. Preservar evid\u00eancias n\u00e3o significa guardar indiscriminadamente tudo o que estiver dispon\u00edvel.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Como estruturar o pr\u00f3ximo passo<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Comece por um caso de uso que influencie diretamente aceita\u00e7\u00e3o, pre\u00e7o ou sinistro. Re\u00fana neg\u00f3cio, jur\u00eddico, riscos e tecnologia para examinar uma decis\u00e3o j\u00e1 tomada.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A partir desse caso, defina as explica\u00e7\u00f5es necess\u00e1rias, seus destinat\u00e1rios e os prazos de resposta. Depois, estabele\u00e7a o que ser\u00e1 preservado, o que poder\u00e1 ser recalculado e como a recupera\u00e7\u00e3o ser\u00e1 testada. Transforme as lacunas encontradas em entregas com respons\u00e1vel e prazo.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Esse trabalho deve fazer parte da <strong><a href=\"https:\/\/tecnologiaunica.com.br\/blog\/governanca-de-ia-em-seguradoras\/\" target=\"_blank\" rel=\"noreferrer noopener\">governan\u00e7a de IA em seguradoras<\/a><\/strong>, com crit\u00e9rios de aprova\u00e7\u00e3o, acompanhamento e revis\u00e3o dos usos.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A configura\u00e7\u00e3o dos produtos e a organiza\u00e7\u00e3o do fluxo de aceita\u00e7\u00e3o s\u00e3o partes dessa arquitetura. <\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Na <strong><a href=\"https:\/\/tecnologiaunica.com.br\/gestao-de-seguros\/\" target=\"_blank\" rel=\"noreferrer noopener\">solu\u00e7\u00e3o de Gest\u00e3o de Seguros da Tecnologia \u00danica<\/a><\/strong>, o m\u00f3dulo administrativo permite configurar produtos, a cota\u00e7\u00e3o utiliza suas parametriza\u00e7\u00f5es e o gerenciamento de workflow organiza a aceita\u00e7\u00e3o. Os controles de explica\u00e7\u00e3o, versionamento e registro por execu\u00e7\u00e3o precisam ser definidos no escopo de cada implementa\u00e7\u00e3o.<\/p>\n\n\n\n<style>\n  .tu-ai-decision-architecture-cta,\n  .tu-ai-decision-architecture-cta * {\n    box-sizing: border-box;\n  }\n\n  .tu-ai-decision-architecture-cta {\n    display: flex;\n    align-items: center;\n    justify-content: space-between;\n    gap: 22px;\n\n    width: 100%;\n    max-width: 100%;\n    margin: 24px 0;\n    padding: 22px 23px;\n\n    background: #000000;\n    border: 2px solid #DD4E4B;\n    border-radius: 4px 16px 16px 16px;\n\n    font-family: \"DM Sans\", Arial, sans-serif;\n  }\n\n  .tu-ai-decision-architecture-cta__content {\n    display: flex;\n    align-items: flex-start;\n    gap: 14px;\n\n    min-width: 0;\n  }\n\n  .tu-ai-decision-architecture-cta__marker {\n    display: flex;\n    align-items: center;\n    justify-content: center;\n\n    flex: 0 0 36px;\n    width: 36px;\n    height: 36px;\n\n    background: #DD4E4B;\n    border-radius: 50%;\n\n    color: #FFFFFF;\n    font-size: 15px;\n    font-weight: 700;\n    line-height: 1;\n  }\n\n  .tu-ai-decision-architecture-cta__text {\n    min-width: 0;\n  }\n\n  .tu-ai-decision-architecture-cta__text strong {\n    display: block;\n    margin: 0 0 6px;\n\n    color: #FFFFFF;\n    font-size: 14.9px;\n    font-weight: 700;\n    line-height: 1.45;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-decision-architecture-cta__text p {\n    margin: 0 !important;\n    padding: 0 !important;\n\n    max-width: 830px;\n\n    color: #E7E7E7 !important;\n    font-size: 13.1px;\n    font-weight: 400;\n    line-height: 1.62;\n\n    word-break: normal !important;\n    overflow-wrap: normal !important;\n    hyphens: none !important;\n  }\n\n  .tu-ai-decision-architecture-cta__text em {\n    color: #FFFFFF;\n    font-style: normal;\n    font-weight: 700;\n  }\n\n  .tu-ai-decision-architecture-cta__button {\n    display: inline-flex;\n    align-items: center;\n    justify-content: center;\n    gap: 8px;\n\n    flex: 0 0 auto;\n\n    min-height: 42px;\n    padding: 10px 17px;\n\n    background: #DD4E4B;\n    border: 1px solid #DD4E4B;\n    border-radius: 8px;\n\n    color: #FFFFFF !important;\n    text-decoration: none !important;\n\n    font-size: 12.6px;\n    font-weight: 700;\n    line-height: 1.2;\n\n    transition:\n      background-color .2s ease,\n      border-color .2s ease,\n      color .2s ease,\n      transform .2s ease;\n  }\n\n  .tu-ai-decision-architecture-cta__button::after {\n    content: \"\u2192\";\n    font-size: 15px;\n    line-height: 1;\n  }\n\n  .tu-ai-decision-architecture-cta__button:hover,\n  .tu-ai-decision-architecture-cta__button:focus {\n    background: #FFFFFF;\n    border-color: #FFFFFF;\n\n    color: #000000 !important;\n    text-decoration: none !important;\n  }\n\n  .tu-ai-decision-architecture-cta__button:focus-visible {\n    outline: 2px solid #FFFFFF;\n    outline-offset: 3px;\n  }\n\n  @media (hover: hover) and (pointer: fine) {\n    .tu-ai-decision-architecture-cta__button:hover {\n      transform: translateY(-1px);\n    }\n  }\n\n  @media (max-width: 680px) {\n    .tu-ai-decision-architecture-cta {\n      flex-direction: column;\n      align-items: flex-start;\n      gap: 16px;\n\n      margin: 20px 0;\n      padding: 19px 18px;\n\n      border-radius: 4px 14px 14px 14px;\n    }\n\n    .tu-ai-decision-architecture-cta__marker {\n      flex-basis: 32px;\n      width: 32px;\n      height: 32px;\n\n      font-size: 14px;\n    }\n\n    .tu-ai-decision-architecture-cta__text strong {\n      font-size: 14.3px;\n    }\n\n    .tu-ai-decision-architecture-cta__text p {\n      font-size: 12.8px;\n    }\n\n    .tu-ai-decision-architecture-cta__button {\n      width: 100%;\n    }\n  }\n<\/style>\n\n<div class=\"tu-ai-decision-architecture-cta\">\n\n  <div class=\"tu-ai-decision-architecture-cta__content\">\n\n    <span\n      class=\"tu-ai-decision-architecture-cta__marker\"\n      aria-hidden=\"true\"\n    >\n      \u2194\n    <\/span>\n\n    <div class=\"tu-ai-decision-architecture-cta__text\">\n\n      <strong>\n        Comece por um \u00fanico processo decis\u00f3rio.\n      <\/strong>\n\n      <p>\n        O primeiro passo pode ser estruturar um fluxo com evid\u00eancias suficientes para\n        <em>compreender o resultado, revisar falhas e sustentar a decis\u00e3o<\/em> antes\n        de ampliar a arquitetura para outros processos da opera\u00e7\u00e3o.\n      <\/p>\n\n    <\/div>\n\n  <\/div>\n\n  <a\n    class=\"tu-ai-decision-architecture-cta__button\"\n    href=\"https:\/\/tecnologiaunica.com.br\/contato\/?utm_source=blog&amp;utm_medium=artigo&amp;utm_content=explicabilidade-ia-seguros\"\n    aria-label=\"Converse com um especialista da Tecnologia \u00danica sobre a arquitetura de decis\u00e3o da sua opera\u00e7\u00e3o\"\n  >\n    Avaliar a arquitetura de decis\u00e3o\n  <\/a>\n\n<\/div>\n\n\n\n<h2 class=\"wp-block-heading\">Perguntas frequentes (FAQ)<\/h2>\n\n\n\n<div class=\"schema-faq wp-block-yoast-faq-block\"><div class=\"schema-faq-section\" id=\"faq-question-1792685286915\"><strong class=\"schema-faq-question\">A LGPD exige a divulga\u00e7\u00e3o do c\u00f3digo do modelo?<\/strong> <p class=\"schema-faq-answer\">O art. 20, \u00a7 1\u00ba, trata de informa\u00e7\u00f5es claras e adequadas sobre crit\u00e9rios e procedimentos, observados os segredos comercial e industrial. N\u00e3o estabelece uma obriga\u00e7\u00e3o geral de entregar c\u00f3digo-fonte. O \u00a7 2\u00ba prev\u00ea a possibilidade de auditoria pela ANPD quando a informa\u00e7\u00e3o for negada com base nesses segredos.<\/p> <\/div> <div class=\"schema-faq-section\" id=\"faq-question-1792685292489\"><strong class=\"schema-faq-question\">SHAP ou LIME bastam para fundamentar uma recusa?<\/strong> <p class=\"schema-faq-answer\">Essas t\u00e9cnicas podem ajudar a analisar a influ\u00eancia das vari\u00e1veis. A fundamenta\u00e7\u00e3o tamb\u00e9m precisa considerar os fatos, as regras aplic\u00e1veis e a forma como o resultado participou da decis\u00e3o. Uma atribui\u00e7\u00e3o t\u00e9cnica n\u00e3o demonstra, por si, que o crit\u00e9rio utilizado era leg\u00edtimo.<\/p> <\/div> <div class=\"schema-faq-section\" id=\"faq-question-1792685300051\"><strong class=\"schema-faq-question\">Quanto tempo os registros devem ser mantidos?<\/strong> <p class=\"schema-faq-answer\">A defini\u00e7\u00e3o deve considerar as exig\u00eancias de conserva\u00e7\u00e3o aplic\u00e1veis, a necessidade de exerc\u00edcio de direitos e a finalidade do tratamento. O prazo e seus fundamentos precisam ser documentados, com participa\u00e7\u00e3o das \u00e1reas respons\u00e1veis e crit\u00e9rios de descarte. N\u00e3o se deve adotar reten\u00e7\u00e3o indefinida apenas porque os registros podem ser \u00fateis no futuro.<\/p> <\/div> <\/div>\n\n\n\n<h2 class=\"wp-block-heading\">Refer\u00eancias<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">[1] <a href=\"https:\/\/www.planalto.gov.br\/ccivil_03\/_ato2015-2018\/2018\/lei\/l13709.htm\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">Planalto. Lei n\u00ba 13.709, de 14 de agosto de 2018 \u2014 Lei Geral de Prote\u00e7\u00e3o de Dados Pessoais (LGPD)<\/a>. <\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[2] <a href=\"https:\/\/www.planalto.gov.br\/ccivil_03\/_ato2023-2026\/2024\/lei\/l15040.htm\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">Planalto. Lei n\u00ba 15.040, de 9 de dezembro de 2024 \u2014 Lei do Contrato de Seguro<\/a>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[3] <a href=\"https:\/\/www2.susep.gov.br\/safe\/scripts\/bnweb\/bnmapi.exe?router=upload\/25061\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">CONSELHO NACIONAL DE SEGUROS PRIVADOS \u2014 CNSP. Resolu\u00e7\u00e3o n\u00ba 416, de 20 de julho de 2021<\/a>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[4] <a href=\"https:\/\/www2.susep.gov.br\/safe\/scripts\/bnweb\/bnmapi.exe?router=upload\/25061\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">Susep. Resolu\u00e7\u00e3o n\u00ba 416, de 20 de julho de 2021<\/a>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[5] <a href=\"https:\/\/www.in.gov.br\/web\/dou\/-\/resolucao-cnsp-n-496-de-17-de-agosto-de-2026-726174976\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">CONSELHO NACIONAL DE SEGUROS PRIVADOS \u2014 CNSP. Resolu\u00e7\u00e3o n\u00ba 496, de 17 de agosto de 2026<\/a>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[6] <a href=\"https:\/\/www.gov.br\/susep\/pt-br\/central-de-conteudos\/noticias\/2026\/agosto\/nova-norma-do-cnsp-estabelece-regras-gerais-para-contratos-de-seguros-de-danos\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">SUPERINTEND\u00caNCIA DE SEGUROS PRIVADOS \u2014 SUSEP. Nova norma do CNSP estabelece regras gerais para contratos de seguros de danos<\/a>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[7] <a href=\"https:\/\/www.gov.br\/susep\/pt-br\/central-de-conteudos\/noticias\/2026\/julho-1\/susep-divulga-manual-de-orientacoes-sobre-seguranca-cibernetica-para-supervisionadas\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">SUPERINTEND\u00caNCIA DE SEGUROS PRIVADOS \u2014 SUSEP. Susep divulga Manual de Orienta\u00e7\u00f5es sobre Seguran\u00e7a Cibern\u00e9tica para supervisionadas<\/a>..<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[8] <a href=\"https:\/\/nvlpubs.nist.gov\/nistpubs\/ir\/2021\/NIST.IR.8312.pdf\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">PHILLIPS, P. Jonathon et al. Four Principles of Explainable Artificial Intelligence<\/a>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[9] <a href=\"https:\/\/arxiv.org\/abs\/1811.10154\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">RUDIN, Cynthia. Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead<\/a>.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[10] <a href=\"https:\/\/shap.readthedocs.io\/en\/latest\/generated\/shap.TreeExplainer.html\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">SHAP. TreeExplainer \u2014 documenta\u00e7\u00e3o t\u00e9cnica<\/a>. <\/p>\n\n\n\n<p class=\"wp-block-paragraph\">[11] <a href=\"https:\/\/pygam.readthedocs.io\/en\/latest\/notebooks\/tour_of_pygam.html\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">PYGAM. A Tour of pyGAM \u2014 documenta\u00e7\u00e3o t\u00e9cnica<\/a>. <\/p>\n","protected":false},"excerpt":{"rendered":"<p>A explicabilidade de IA em seguros ajuda a compreender e revisar decis\u00f5es sobre aceita\u00e7\u00e3o, pre\u00e7os e sinistros, com base em crit\u00e9rios t\u00e9cnicos, fundamenta\u00e7\u00e3o, rastreabilidade e controles operacionais.<\/p>","protected":false},"author":10,"featured_media":6393,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[30],"tags":[],"class_list":["post-6300","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tecnologia-para-seguros"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.1 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Explicabilidade de IA em seguros: como sustentar cada decis\u00e3o<\/title>\n<meta name=\"description\" content=\"Entenda a explicabilidade de IA em seguros, os deveres e como preservar evid\u00eancias e fundamentar decis\u00f5es sobre propostas e sinistros.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/tecnologiaunica.com.br\/it\/blog\/explicabilidade-de-ia-em-seguros\/\" \/>\n<meta property=\"og:locale\" content=\"it_IT\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Explicabilidade de IA em seguros: como sustentar cada decis\u00e3o\" \/>\n<meta property=\"og:description\" content=\"Entenda a explicabilidade de IA em seguros, os deveres e como preservar evid\u00eancias e fundamentar decis\u00f5es sobre propostas e sinistros.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/tecnologiaunica.com.br\/it\/blog\/explicabilidade-de-ia-em-seguros\/\" \/>\n<meta property=\"og:site_name\" content=\"Tecnologia \u00danica\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-22T16:16:11+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-10-23T15:54:56+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/tecnologiaunica.com.br\/wp-content\/uploads\/2026\/10\/explicabilidade-de-ia-em-seguros-1.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1200\" \/>\n\t<meta property=\"og:image:height\" content=\"675\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\n<meta name=\"author\" content=\"Isabelle Fujioka\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Scritto da\" \/>\n\t<meta name=\"twitter:data1\" content=\"Isabelle Fujioka\" \/>\n\t<meta name=\"twitter:label2\" content=\"Tempo di lettura stimato\" \/>\n\t<meta name=\"twitter:data2\" content=\"15 minuti\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/\"},\"author\":{\"name\":\"Isabelle Fujioka\",\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/#\\\/schema\\\/person\\\/2e62384dd46d728df9965ca9dc78cb34\"},\"headline\":\"Explicabilidade de IA em seguros: como sustentar cada decis\u00e3o\",\"datePublished\":\"2026-10-22T16:16:11+00:00\",\"dateModified\":\"2026-10-23T15:54:56+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/\"},\"wordCount\":2974,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/explicabilidade-de-ia-em-seguros-1.webp\",\"articleSection\":[\"Tecnologia para Seguros\"],\"inLanguage\":\"it-IT\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#respond\"]}]},{\"@type\":[\"WebPage\",\"FAQPage\"],\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/\",\"url\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/\",\"name\":\"Explicabilidade de IA em seguros: como sustentar cada decis\u00e3o\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/explicabilidade-de-ia-em-seguros-1.webp\",\"datePublished\":\"2026-10-22T16:16:11+00:00\",\"dateModified\":\"2026-10-23T15:54:56+00:00\",\"description\":\"Entenda a explicabilidade de IA em seguros, os deveres e como preservar evid\u00eancias e fundamentar decis\u00f5es sobre propostas e sinistros.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#breadcrumb\"},\"mainEntity\":[{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#faq-question-1792685286915\"},{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#faq-question-1792685292489\"},{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#faq-question-1792685300051\"}],\"inLanguage\":\"it-IT\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"it-IT\",\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#primaryimage\",\"url\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/explicabilidade-de-ia-em-seguros-1.webp\",\"contentUrl\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/explicabilidade-de-ia-em-seguros-1.webp\",\"width\":1200,\"height\":675,\"caption\":\"Equipe em escrit\u00f3rio analisa relat\u00f3rios e gr\u00e1ficos em telas de computador, enquanto uma mulher aponta dados em documentos durante uma reuni\u00e3o corporativa, contexto para explicabilidade de IA em seguros.\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"In\u00edcio\",\"item\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Explicabilidade de IA em seguros: como sustentar cada decis\u00e3o\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/#website\",\"url\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/\",\"name\":\"Tecnologia \u00danica\",\"description\":\"IA, outsourcing, marketplace e muito mais\",\"publisher\":{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/#organization\"},\"alternateName\":\"Tecnologia \u00danica\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"it-IT\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/#organization\",\"name\":\"Tecnologia \u00danica\",\"alternateName\":\"Tecnologia \u00danica\",\"url\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"it-IT\",\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/wp-content\\\/uploads\\\/2023\\\/08\\\/tecnologia_unica_logo__original-1.png\",\"contentUrl\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/wp-content\\\/uploads\\\/2023\\\/08\\\/tecnologia_unica_logo__original-1.png\",\"width\":213,\"height\":35,\"caption\":\"Tecnologia \u00danica\"},\"image\":{\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/#\\\/schema\\\/logo\\\/image\\\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/#\\\/schema\\\/person\\\/2e62384dd46d728df9965ca9dc78cb34\",\"name\":\"Isabelle Fujioka\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"it-IT\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/d75d40a3e1c9087f876dd0670f79a751670b0da7071720d3b4c1bedc95f08c28?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/d75d40a3e1c9087f876dd0670f79a751670b0da7071720d3b4c1bedc95f08c28?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/d75d40a3e1c9087f876dd0670f79a751670b0da7071720d3b4c1bedc95f08c28?s=96&d=mm&r=g\",\"caption\":\"Isabelle Fujioka\"},\"url\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/it\\\/blog\\\/author\\\/ifujioka\\\/\"},{\"@type\":\"Question\",\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#faq-question-1792685286915\",\"position\":1,\"url\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#faq-question-1792685286915\",\"name\":\"A LGPD exige a divulga\u00e7\u00e3o do c\u00f3digo do modelo?\",\"answerCount\":1,\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"O art. 20, \u00a7 1\u00ba, trata de informa\u00e7\u00f5es claras e adequadas sobre crit\u00e9rios e procedimentos, observados os segredos comercial e industrial. N\u00e3o estabelece uma obriga\u00e7\u00e3o geral de entregar c\u00f3digo-fonte. O \u00a7 2\u00ba prev\u00ea a possibilidade de auditoria pela ANPD quando a informa\u00e7\u00e3o for negada com base nesses segredos.\",\"inLanguage\":\"it-IT\"},\"inLanguage\":\"it-IT\"},{\"@type\":\"Question\",\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#faq-question-1792685292489\",\"position\":2,\"url\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#faq-question-1792685292489\",\"name\":\"SHAP ou LIME bastam para fundamentar uma recusa?\",\"answerCount\":1,\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"Essas t\u00e9cnicas podem ajudar a analisar a influ\u00eancia das vari\u00e1veis. A fundamenta\u00e7\u00e3o tamb\u00e9m precisa considerar os fatos, as regras aplic\u00e1veis e a forma como o resultado participou da decis\u00e3o. Uma atribui\u00e7\u00e3o t\u00e9cnica n\u00e3o demonstra, por si, que o crit\u00e9rio utilizado era leg\u00edtimo.\",\"inLanguage\":\"it-IT\"},\"inLanguage\":\"it-IT\"},{\"@type\":\"Question\",\"@id\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#faq-question-1792685300051\",\"position\":3,\"url\":\"https:\\\/\\\/tecnologiaunica.com.br\\\/blog\\\/explicabilidade-de-ia-em-seguros\\\/#faq-question-1792685300051\",\"name\":\"Quanto tempo os registros devem ser mantidos?\",\"answerCount\":1,\"acceptedAnswer\":{\"@type\":\"Answer\",\"text\":\"A defini\u00e7\u00e3o deve considerar as exig\u00eancias de conserva\u00e7\u00e3o aplic\u00e1veis, a necessidade de exerc\u00edcio de direitos e a finalidade do tratamento. O prazo e seus fundamentos precisam ser documentados, com participa\u00e7\u00e3o das \u00e1reas respons\u00e1veis e crit\u00e9rios de descarte. N\u00e3o se deve adotar reten\u00e7\u00e3o indefinida apenas porque os registros podem ser \u00fateis no futuro.\",\"inLanguage\":\"it-IT\"},\"inLanguage\":\"it-IT\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Explicabilidade de IA em seguros: como sustentar cada decis\u00e3o","description":"Entenda a explicabilidade de IA em seguros, os deveres e como preservar evid\u00eancias e fundamentar decis\u00f5es sobre propostas e sinistros.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/tecnologiaunica.com.br\/it\/blog\/explicabilidade-de-ia-em-seguros\/","og_locale":"it_IT","og_type":"article","og_title":"Explicabilidade de IA em seguros: como sustentar cada decis\u00e3o","og_description":"Entenda a explicabilidade de IA em seguros, os deveres e como preservar evid\u00eancias e fundamentar decis\u00f5es sobre propostas e sinistros.","og_url":"https:\/\/tecnologiaunica.com.br\/it\/blog\/explicabilidade-de-ia-em-seguros\/","og_site_name":"Tecnologia \u00danica","article_published_time":"2026-10-22T16:16:11+00:00","article_modified_time":"2026-10-23T15:54:56+00:00","og_image":[{"width":1200,"height":675,"url":"https:\/\/tecnologiaunica.com.br\/wp-content\/uploads\/2026\/10\/explicabilidade-de-ia-em-seguros-1.webp","type":"image\/webp"}],"author":"Isabelle Fujioka","twitter_card":"summary_large_image","twitter_misc":{"Scritto da":"Isabelle Fujioka","Tempo di lettura stimato":"15 minuti"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#article","isPartOf":{"@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/"},"author":{"name":"Isabelle Fujioka","@id":"https:\/\/tecnologiaunica.com.br\/#\/schema\/person\/2e62384dd46d728df9965ca9dc78cb34"},"headline":"Explicabilidade de IA em seguros: como sustentar cada decis\u00e3o","datePublished":"2026-10-22T16:16:11+00:00","dateModified":"2026-10-23T15:54:56+00:00","mainEntityOfPage":{"@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/"},"wordCount":2974,"commentCount":0,"publisher":{"@id":"https:\/\/tecnologiaunica.com.br\/#organization"},"image":{"@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#primaryimage"},"thumbnailUrl":"https:\/\/tecnologiaunica.com.br\/wp-content\/uploads\/2026\/10\/explicabilidade-de-ia-em-seguros-1.webp","articleSection":["Tecnologia para Seguros"],"inLanguage":"it-IT","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#respond"]}]},{"@type":["WebPage","FAQPage"],"@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/","url":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/","name":"Explicabilidade de IA em seguros: como sustentar cada decis\u00e3o","isPartOf":{"@id":"https:\/\/tecnologiaunica.com.br\/#website"},"primaryImageOfPage":{"@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#primaryimage"},"image":{"@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#primaryimage"},"thumbnailUrl":"https:\/\/tecnologiaunica.com.br\/wp-content\/uploads\/2026\/10\/explicabilidade-de-ia-em-seguros-1.webp","datePublished":"2026-10-22T16:16:11+00:00","dateModified":"2026-10-23T15:54:56+00:00","description":"Entenda a explicabilidade de IA em seguros, os deveres e como preservar evid\u00eancias e fundamentar decis\u00f5es sobre propostas e sinistros.","breadcrumb":{"@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#breadcrumb"},"mainEntity":[{"@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#faq-question-1792685286915"},{"@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#faq-question-1792685292489"},{"@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#faq-question-1792685300051"}],"inLanguage":"it-IT","potentialAction":[{"@type":"ReadAction","target":["https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/"]}]},{"@type":"ImageObject","inLanguage":"it-IT","@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#primaryimage","url":"https:\/\/tecnologiaunica.com.br\/wp-content\/uploads\/2026\/10\/explicabilidade-de-ia-em-seguros-1.webp","contentUrl":"https:\/\/tecnologiaunica.com.br\/wp-content\/uploads\/2026\/10\/explicabilidade-de-ia-em-seguros-1.webp","width":1200,"height":675,"caption":"Equipe em escrit\u00f3rio analisa relat\u00f3rios e gr\u00e1ficos em telas de computador, enquanto uma mulher aponta dados em documentos durante uma reuni\u00e3o corporativa, contexto para explicabilidade de IA em seguros."},{"@type":"BreadcrumbList","@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"In\u00edcio","item":"https:\/\/tecnologiaunica.com.br\/"},{"@type":"ListItem","position":2,"name":"Explicabilidade de IA em seguros: como sustentar cada decis\u00e3o"}]},{"@type":"WebSite","@id":"https:\/\/tecnologiaunica.com.br\/#website","url":"https:\/\/tecnologiaunica.com.br\/","name":"Tecnologia \u00danica","description":"IA, outsourcing, marketplace e muito mais","publisher":{"@id":"https:\/\/tecnologiaunica.com.br\/#organization"},"alternateName":"Tecnologia \u00danica","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/tecnologiaunica.com.br\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"it-IT"},{"@type":"Organization","@id":"https:\/\/tecnologiaunica.com.br\/#organization","name":"Tecnologia \u00danica","alternateName":"Tecnologia \u00danica","url":"https:\/\/tecnologiaunica.com.br\/","logo":{"@type":"ImageObject","inLanguage":"it-IT","@id":"https:\/\/tecnologiaunica.com.br\/#\/schema\/logo\/image\/","url":"https:\/\/tecnologiaunica.com.br\/wp-content\/uploads\/2023\/08\/tecnologia_unica_logo__original-1.png","contentUrl":"https:\/\/tecnologiaunica.com.br\/wp-content\/uploads\/2023\/08\/tecnologia_unica_logo__original-1.png","width":213,"height":35,"caption":"Tecnologia \u00danica"},"image":{"@id":"https:\/\/tecnologiaunica.com.br\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/tecnologiaunica.com.br\/#\/schema\/person\/2e62384dd46d728df9965ca9dc78cb34","name":"Isabelle Fujioka","image":{"@type":"ImageObject","inLanguage":"it-IT","@id":"https:\/\/secure.gravatar.com\/avatar\/d75d40a3e1c9087f876dd0670f79a751670b0da7071720d3b4c1bedc95f08c28?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/d75d40a3e1c9087f876dd0670f79a751670b0da7071720d3b4c1bedc95f08c28?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/d75d40a3e1c9087f876dd0670f79a751670b0da7071720d3b4c1bedc95f08c28?s=96&d=mm&r=g","caption":"Isabelle Fujioka"},"url":"https:\/\/tecnologiaunica.com.br\/it\/blog\/author\/ifujioka\/"},{"@type":"Question","@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#faq-question-1792685286915","position":1,"url":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#faq-question-1792685286915","name":"A LGPD exige a divulga\u00e7\u00e3o do c\u00f3digo do modelo?","answerCount":1,"acceptedAnswer":{"@type":"Answer","text":"O art. 20, \u00a7 1\u00ba, trata de informa\u00e7\u00f5es claras e adequadas sobre crit\u00e9rios e procedimentos, observados os segredos comercial e industrial. N\u00e3o estabelece uma obriga\u00e7\u00e3o geral de entregar c\u00f3digo-fonte. O \u00a7 2\u00ba prev\u00ea a possibilidade de auditoria pela ANPD quando a informa\u00e7\u00e3o for negada com base nesses segredos.","inLanguage":"it-IT"},"inLanguage":"it-IT"},{"@type":"Question","@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#faq-question-1792685292489","position":2,"url":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#faq-question-1792685292489","name":"SHAP ou LIME bastam para fundamentar uma recusa?","answerCount":1,"acceptedAnswer":{"@type":"Answer","text":"Essas t\u00e9cnicas podem ajudar a analisar a influ\u00eancia das vari\u00e1veis. A fundamenta\u00e7\u00e3o tamb\u00e9m precisa considerar os fatos, as regras aplic\u00e1veis e a forma como o resultado participou da decis\u00e3o. Uma atribui\u00e7\u00e3o t\u00e9cnica n\u00e3o demonstra, por si, que o crit\u00e9rio utilizado era leg\u00edtimo.","inLanguage":"it-IT"},"inLanguage":"it-IT"},{"@type":"Question","@id":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#faq-question-1792685300051","position":3,"url":"https:\/\/tecnologiaunica.com.br\/blog\/explicabilidade-de-ia-em-seguros\/#faq-question-1792685300051","name":"Quanto tempo os registros devem ser mantidos?","answerCount":1,"acceptedAnswer":{"@type":"Answer","text":"A defini\u00e7\u00e3o deve considerar as exig\u00eancias de conserva\u00e7\u00e3o aplic\u00e1veis, a necessidade de exerc\u00edcio de direitos e a finalidade do tratamento. O prazo e seus fundamentos precisam ser documentados, com participa\u00e7\u00e3o das \u00e1reas respons\u00e1veis e crit\u00e9rios de descarte. N\u00e3o se deve adotar reten\u00e7\u00e3o indefinida apenas porque os registros podem ser \u00fateis no futuro.","inLanguage":"it-IT"},"inLanguage":"it-IT"}]}},"_links":{"self":[{"href":"https:\/\/tecnologiaunica.com.br\/it\/wp-json\/wp\/v2\/posts\/6300","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/tecnologiaunica.com.br\/it\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/tecnologiaunica.com.br\/it\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/tecnologiaunica.com.br\/it\/wp-json\/wp\/v2\/users\/10"}],"replies":[{"embeddable":true,"href":"https:\/\/tecnologiaunica.com.br\/it\/wp-json\/wp\/v2\/comments?post=6300"}],"version-history":[{"count":1,"href":"https:\/\/tecnologiaunica.com.br\/it\/wp-json\/wp\/v2\/posts\/6300\/revisions"}],"predecessor-version":[{"id":6302,"href":"https:\/\/tecnologiaunica.com.br\/it\/wp-json\/wp\/v2\/posts\/6300\/revisions\/6302"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/tecnologiaunica.com.br\/it\/wp-json\/wp\/v2\/media\/6393"}],"wp:attachment":[{"href":"https:\/\/tecnologiaunica.com.br\/it\/wp-json\/wp\/v2\/media?parent=6300"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/tecnologiaunica.com.br\/it\/wp-json\/wp\/v2\/categories?post=6300"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/tecnologiaunica.com.br\/it\/wp-json\/wp\/v2\/tags?post=6300"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}