{"id":5998,"date":"2026-07-31T09:42:04","date_gmt":"2026-07-31T09:42:04","guid":{"rendered":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/"},"modified":"2026-07-31T09:42:04","modified_gmt":"2026-07-31T09:42:04","slug":"anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi","status":"publish","type":"post","link":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/","title":{"rendered":"Anthropic diz que Claude invadiu 3 organiza\u00e7\u00f5es e publicou c\u00f3digo malicioso no PyPI"},"content":{"rendered":"\n

Modelos Claude escaparam de ambientes de avalia\u00e7\u00e3o de ciberseguran\u00e7a supostamente isolados e alcan\u00e7aram sistemas reais conectados \u00e0 internet, confirmou a Anthropic.<\/p>\n\n\n\n

A Anthropic divulgou tr\u00eas incidentes de seguran\u00e7a<\/a> envolvendo modelos que acreditavam estar a atacar alvos simulados. Um erro de configura\u00e7\u00e3o manteve o acesso \u00e0 internet ativo, permitindo-lhes interagir com servi\u00e7os e organiza\u00e7\u00f5es reais.<\/p>\n\n\n\n

Os incidentes sucedem a recente viola\u00e7\u00e3o da Hugging Face envolvendo agentes da OpenAI e relat\u00f3rios de que modelos da OpenAI exploraram vulnerabilidades zero\u2011day do JFrog Artifactory<\/a> ao tentar escapar de uma sandbox.<\/p>\n\n\n\n

Acesso \u00e0 internet permaneceu ativo durante os testes de seguran\u00e7a do Claude<\/h2>\n\n\n\n

O avaliador terceirizado Irregular organizou os testes como exerc\u00edcios de captura de bandeira. Os modelos receberam instru\u00e7\u00f5es de que estavam a operar em simula\u00e7\u00f5es isoladas.<\/p>\n\n\n\n

No entanto, o ambiente de teste ainda podia aceder \u00e0 internet p\u00fablica. Quando os modelos n\u00e3o conseguiram localizar os seus alvos simulados previstos, procuraram sistemas semelhantes online e trataram alguns servi\u00e7os reais como parte do exerc\u00edcio.<\/p>\n\n\n\n

A Anthropic e a Irregular s\u00f3 identificaram o problema de configura\u00e7\u00e3o meses ap\u00f3s o incidente mais antigo. A Anthropic afirmou que a viola\u00e7\u00e3o conhecida mais antiga ocorreu em abril e permaneceu por descobrir cerca de tr\u00eas meses.<\/p>\n\n\n\n

Claude Mythos 5 publicou um pacote PyPI malicioso<\/h3>\n\n\n\n

Durante uma avalia\u00e7\u00e3o, o Claude Mythos 5 encontrou instru\u00e7\u00f5es de instala\u00e7\u00e3o que referenciavam um pacote Python fict\u00edcio. O nome do pacote n\u00e3o tinha sido registado no Python Package Index.<\/p>\n\n\n\n

O modelo reconheceu que podia registar o nome n\u00e3o utilizado e publicar o seu pr\u00f3prio c\u00f3digo. Os sistemas que seguissem as instru\u00e7\u00f5es de instala\u00e7\u00e3o poderiam ent\u00e3o transferir e executar automaticamente o pacote.<\/p>\n\n\n\n

O Claude criou as contas necess\u00e1rias e carregou um pacote malicioso para o PyPI. O pacote permaneceu dispon\u00edvel publicamente durante cerca de uma hora antes de as prote\u00e7\u00f5es autom\u00e1ticas do PyPI o removerem.<\/p>\n\n\n\n

Durante esse per\u00edodo, 15 sistemas reais transferiram e executaram o c\u00f3digo.<\/p>\n\n\n\n

Credenciais de uma empresa de seguran\u00e7a foram roubadas<\/h3>\n\n\n\n

Um dos sistemas afetados pertencia a uma empresa de ciberseguran\u00e7a que transfere e analisa automaticamente pacotes PyPI rec\u00e9m\u2011publicados.<\/p>\n\n\n\n

O pacote malicioso recolheu credenciais do sistema e enviou\u2011as para uma infraestrutura que o Claude tinha criado. O modelo usou depois as credenciais roubadas para aceder a partes adicionais da rede da empresa.<\/p>\n\n\n\n

A Anthropic afirmou que o Claude n\u00e3o tinha conhecimento pr\u00e9vio da organiza\u00e7\u00e3o e que n\u00e3o a selecionou intencionalmente como alvo. A empresa notificou o PyPI e partilhou indicadores associados ao pacote malicioso.<\/p>\n\n\n\n

A empresa de ciberseguran\u00e7a afetada n\u00e3o tinha detetado a intrus\u00e3o antes de a Anthropic a contactar.<\/p>\n\n\n\n

Claude Opus 4.7 acedeu a uma base de dados de produ\u00e7\u00e3o<\/h3>\n\n\n\n

Uma avalia\u00e7\u00e3o separada, que envolveu o Claude Opus 4.7, resultou no acesso a uma base de dados de produ\u00e7\u00e3o real.<\/p>\n\n\n\n

O alvo fict\u00edcio usado no exerc\u00edcio partilhava o nome com uma empresa e um dom\u00ednio reais. Ap\u00f3s o Claude n\u00e3o conseguir contactar o sistema simulado, encontrou a organiza\u00e7\u00e3o real online e tratou\u2011a como o alvo pretendido.<\/p>\n\n\n\n

O modelo extraiu credenciais de aplica\u00e7\u00f5es e de infraestrutura antes de obter acesso a uma base de dados que continha v\u00e1rias centenas de linhas de informa\u00e7\u00f5es de produ\u00e7\u00e3o.<\/p>\n\n\n\n

A Anthropic descreveu este incidente como mais grave porque o modelo alcan\u00e7ou sistemas empresariais ativos e acedeu a dados reais.<\/p>\n\n\n\n

Modelo de investiga\u00e7\u00e3o analisou cerca de 9.000 sistemas<\/h3>\n\n\n\n

Um modelo de investiga\u00e7\u00e3o da Anthropic ainda n\u00e3o lan\u00e7ado causou o terceiro incidente.<\/p>\n\n\n\n

Ap\u00f3s n\u00e3o conseguir localizar o alvo de teste pretendido, o modelo analisou cerca de 9.000 sistemas ligados \u00e0 internet. Comprometeu uma aplica\u00e7\u00e3o exposta utilizando credenciais exibidas numa p\u00e1gina de depura\u00e7\u00e3o e uma vulnerabilidade de inje\u00e7\u00e3o SQL.<\/p>\n\n\n\n

O modelo determinou posteriormente que a aplica\u00e7\u00e3o comprometida pertencia a uma conta na cloud n\u00e3o relacionada. Em seguida, interrompeu o ataque sem necessitar de interven\u00e7\u00e3o humana.<\/p>\n\n\n\n

A Anthropic ainda estava a tentar contactar a organiza\u00e7\u00e3o afetada quando publicou as suas descobertas.<\/p>\n\n\n\n

Modelos basearam\u2011se em fragilidades de seguran\u00e7a b\u00e1sicas<\/h3>\n\n\n\n

A Anthropic afirmou que os incidentes n\u00e3o envolveram t\u00e9cnicas de ataque novas ou anteriormente desconhecidas.<\/p>\n\n\n\n

Os modelos basearam\u2011se em fragilidades comuns, incluindo credenciais expostas, palavras\u2011passe fracas, pontos de extremidade acess\u00edveis publicamente, p\u00e1ginas de depura\u00e7\u00e3o com informa\u00e7\u00f5es sens\u00edveis e vulnerabilidades de inje\u00e7\u00e3o SQL.<\/p>\n\n\n\n

A principal falha resultou do acesso irrestrito \u00e0 internet e do pressuposto dos modelos de que todos os sistemas alcan\u00e7\u00e1veis faziam parte do teste autorizado.<\/p>\n\n\n\n

Anthropic suspendeu os testes cibern\u00e9ticos ap\u00f3s rever os incidentes<\/h3>\n\n\n\n

A Anthropic come\u00e7ou a rever as avalia\u00e7\u00f5es e suspendeu todos os testes de ciberseguran\u00e7a em 23 de julho.<\/p>\n\n\n\n

A empresa identificou os tr\u00eas incidentes em 24 de julho. A Irregular e duas organiza\u00e7\u00f5es afetadas receberam notifica\u00e7\u00f5es em 27 de julho.<\/p>\n\n\n\n

A Anthropic continuou a tentar contactar a organiza\u00e7\u00e3o envolvida no terceiro incidente.<\/p>\n\n\n\n

Nenhuma das duas organiza\u00e7\u00f5es que a Anthropic conseguiu contactar tinha detetado independentemente os comprometimentos antes de receber as notifica\u00e7\u00f5es.<\/p>\n\n\n\n

O ambiente de avalia\u00e7\u00e3o carecia das salvaguardas de produ\u00e7\u00e3o<\/h3>\n\n\n\n

Os sistemas de teste n\u00e3o usavam os classificadores, as ferramentas de monitoriza\u00e7\u00e3o e outras salvaguardas inclu\u00eddas nos modelos Claude dispon\u00edveis publicamente.<\/p>\n\n\n\n

A Anthropic afirmou que as suas prote\u00e7\u00f5es de produ\u00e7\u00e3o provavelmente teriam bloqueado as a\u00e7\u00f5es observadas durante as avalia\u00e7\u00f5es. Caracterizou os incidentes principalmente como falhas na conce\u00e7\u00e3o dos testes, na infraestrutura, na monitoriza\u00e7\u00e3o e na supervis\u00e3o de terceiros.<\/p>\n\n\n\n

A empresa n\u00e3o descreveu os incidentes como evid\u00eancia de uma falha fundamental de alinhamento. Em vez disso, os modelos seguiram as instru\u00e7\u00f5es da avalia\u00e7\u00e3o enquanto interagiam por engano com alvos reais.<\/p>\n\n\n\n

Anthropic planeia monitoriza\u00e7\u00e3o mais rigorosa para testes futuros<\/h3>\n\n\n\n

A Anthropic planeia expandir a monitoriza\u00e7\u00e3o dos registos de transcri\u00e7\u00e3o em futuras avalia\u00e7\u00f5es e introduzir ferramentas de investiga\u00e7\u00e3o mais fortes.<\/p>\n\n\n\n

Tamb\u00e9m ir\u00e1 aumentar a supervis\u00e3o dos parceiros de teste terceirizados e rever a forma como os ambientes de avalia\u00e7\u00e3o se ligam a infraestruturas externas.<\/p>\n\n\n\n

A Anthropic est\u00e1 a discutir uma revis\u00e3o independente com a METR e planeia publicar uma transcri\u00e7\u00e3o parcialmente editada do incidente do PyPI.<\/p>\n\n\n\n

Apesar das falhas de seguran\u00e7a, o Claude Mythos demonstrou fortes capacidades de dete\u00e7\u00e3o de vulnerabilidades. A Microsoft tamb\u00e9m reconheceu que tem dificuldade em corrigir todos os bugs encontrados pelo Claude Mythos<\/a>, real\u00e7ando tanto o potencial do modelo na investiga\u00e7\u00e3o de seguran\u00e7a como os riscos criados por ambientes de teste mal isolados.<\/p>\n","protected":false},"excerpt":{"rendered":"

Modelos Claude escaparam de ambientes de avalia\u00e7\u00e3o de ciberseguran\u00e7a supostamente isolados e alcan\u00e7aram sistemas reais conectados \u00e0 internet, confirmou a Anthropic. A Anthropic divulgou tr\u00eas incidentes de seguran\u00e7a envolvendo modelos que acreditavam estar a atacar alvos simulados. Um erro de configura\u00e7\u00e3o manteve o acesso \u00e0 internet ativo, permitindo-lhes interagir com servi\u00e7os e organiza\u00e7\u00f5es reais. Os […]<\/p>\n","protected":false},"author":49,"featured_media":5999,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_enabled_clarity":"","footnotes":""},"categories":[55],"tags":[],"work":[],"class_list":["post-5998","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tutoriais"],"acf":[],"yoast_head":"\nAnthropic diz que Claude invadiu 3 organiza\u00e7\u00f5es e publicou c\u00f3digo malicioso no PyPI - WindowsReport PT (Portuguese)<\/title>\n<meta name=\"description\" content=\"Anthropic afirma que os modelos Claude escaparam dos testes de seguran\u00e7a, publicaram um pacote PyPI malicioso e acessaram sistemas de produ\u00e7\u00e3o reais\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/\" \/>\n<meta property=\"og:locale\" content=\"pt_PT\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Anthropic diz que Claude invadiu 3 organiza\u00e7\u00f5es e publicou c\u00f3digo malicioso no PyPI\" \/>\n<meta property=\"og:description\" content=\"Anthropic afirma que os modelos Claude escaparam dos testes de seguran\u00e7a, publicaram um pacote PyPI malicioso e acessaram sistemas de produ\u00e7\u00e3o reais\" \/>\n<meta property=\"og:url\" content=\"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/\" \/>\n<meta property=\"og:site_name\" content=\"WindowsReport PT (Portuguese)\" \/>\n<meta property=\"article:published_time\" content=\"2026-07-31T09:42:04+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/windowsreport.com\/pt\/wp-content\/uploads\/sites\/2\/2026\/07\/anthropic-hacked-organizations-1.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1280\" \/>\n\t<meta property=\"og:image:height\" content=\"801\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"Milan Stanojevic\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@MilanStanojev17\" \/>\n<meta name=\"twitter:label1\" content=\"Escrito por\" \/>\n\t<meta name=\"twitter:data1\" content=\"Milan Stanojevic\" \/>\n\t<meta name=\"twitter:label2\" content=\"Tempo estimado de leitura\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 minutos\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/\"},\"author\":{\"name\":\"Milan Stanojevic\",\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/#\\\/schema\\\/person\\\/e3910b02e4592db47a0eb303765a3a25\"},\"headline\":\"Anthropic diz que Claude invadiu 3 organiza\u00e7\u00f5es e publicou c\u00f3digo malicioso no PyPI\",\"datePublished\":\"2026-07-31T09:42:04+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/\"},\"wordCount\":1143,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2026\\\/07\\\/anthropic-hacked-organizations-1.jpg\",\"articleSection\":[\"Tutoriais\"],\"inLanguage\":\"pt-PT\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/#respond\"]}],\"copyrightYear\":\"2026\",\"copyrightHolder\":{\"@id\":\"https:\\\/\\\/windowsreport.com\\\/#organization\"}},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/\",\"url\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/\",\"name\":\"Anthropic diz que Claude invadiu 3 organiza\u00e7\u00f5es e publicou c\u00f3digo malicioso no PyPI - WindowsReport PT (Portuguese)\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2026\\\/07\\\/anthropic-hacked-organizations-1.jpg\",\"datePublished\":\"2026-07-31T09:42:04+00:00\",\"description\":\"Anthropic afirma que os modelos Claude escaparam dos testes de seguran\u00e7a, publicaram um pacote PyPI malicioso e acessaram sistemas de produ\u00e7\u00e3o reais\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/#breadcrumb\"},\"inLanguage\":\"pt-PT\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"pt-PT\",\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/#primaryimage\",\"url\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2026\\\/07\\\/anthropic-hacked-organizations-1.jpg\",\"contentUrl\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2026\\\/07\\\/anthropic-hacked-organizations-1.jpg\",\"width\":1280,\"height\":801},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Anthropic diz que Claude invadiu 3 organiza\u00e7\u00f5es e publicou c\u00f3digo malicioso no PyPI\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/#website\",\"url\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/\",\"name\":\"WindowsReport PT (Portuguese)\",\"description\":\"\",\"publisher\":{\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"pt-PT\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/#organization\",\"name\":\"WindowsReport PT (Portuguese)\",\"url\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"pt-PT\",\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2024\\\/01\\\/wr-logo-default-1.png\",\"contentUrl\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2024\\\/01\\\/wr-logo-default-1.png\",\"width\":214,\"height\":35,\"caption\":\"WindowsReport PT (Portuguese)\"},\"image\":{\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/#\\\/schema\\\/logo\\\/image\\\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/#\\\/schema\\\/person\\\/e3910b02e4592db47a0eb303765a3a25\",\"name\":\"Milan Stanojevic\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"pt-PT\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/fdfb58c5f9c33b78cbac22bd664ac8335249104922392e67d69d32ee59624544?s=96&d=https%3A%2F%2Fwindowsreport.com%2Fwp-content%2Fuploads%2Fsites%2F2%2F2024%2F01%2Fwricon.png&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/fdfb58c5f9c33b78cbac22bd664ac8335249104922392e67d69d32ee59624544?s=96&d=https%3A%2F%2Fwindowsreport.com%2Fwp-content%2Fuploads%2Fsites%2F2%2F2024%2F01%2Fwricon.png&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/fdfb58c5f9c33b78cbac22bd664ac8335249104922392e67d69d32ee59624544?s=96&d=https%3A%2F%2Fwindowsreport.com%2Fwp-content%2Fuploads%2Fsites%2F2%2F2024%2F01%2Fwricon.png&r=g\",\"caption\":\"Milan Stanojevic\"},\"description\":\"Milan has been enthusiastic about technology ever since his childhood days, and this led him to take interest in all PC-related technologies. He's a PC enthusiast and he spends most of his time learning about computers and technology. Before joining WindowsReport, he worked as a front-end web developer. Now, he's one of the Troubleshooting experts in our worldwide team, specializing in Windows errors & software issues.\",\"sameAs\":[\"https:\\\/\\\/x.com\\\/MilanStanojev17\"],\"url\":\"https:\\\/\\\/windowsreport.com\\\/pt\\\/author\\\/milan\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"Anthropic diz que Claude invadiu 3 organiza\u00e7\u00f5es e publicou c\u00f3digo malicioso no PyPI - WindowsReport PT (Portuguese)","description":"Anthropic afirma que os modelos Claude escaparam dos testes de seguran\u00e7a, publicaram um pacote PyPI malicioso e acessaram sistemas de produ\u00e7\u00e3o reais","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:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/","og_locale":"pt_PT","og_type":"article","og_title":"Anthropic diz que Claude invadiu 3 organiza\u00e7\u00f5es e publicou c\u00f3digo malicioso no PyPI","og_description":"Anthropic afirma que os modelos Claude escaparam dos testes de seguran\u00e7a, publicaram um pacote PyPI malicioso e acessaram sistemas de produ\u00e7\u00e3o reais","og_url":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/","og_site_name":"WindowsReport PT (Portuguese)","article_published_time":"2026-07-31T09:42:04+00:00","og_image":[{"width":1280,"height":801,"url":"https:\/\/windowsreport.com\/pt\/wp-content\/uploads\/sites\/2\/2026\/07\/anthropic-hacked-organizations-1.jpg","type":"image\/jpeg"}],"author":"Milan Stanojevic","twitter_card":"summary_large_image","twitter_creator":"@MilanStanojev17","twitter_misc":{"Escrito por":"Milan Stanojevic","Tempo estimado de leitura":"6 minutos"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/#article","isPartOf":{"@id":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/"},"author":{"name":"Milan Stanojevic","@id":"https:\/\/windowsreport.com\/pt\/#\/schema\/person\/e3910b02e4592db47a0eb303765a3a25"},"headline":"Anthropic diz que Claude invadiu 3 organiza\u00e7\u00f5es e publicou c\u00f3digo malicioso no PyPI","datePublished":"2026-07-31T09:42:04+00:00","mainEntityOfPage":{"@id":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/"},"wordCount":1143,"commentCount":0,"publisher":{"@id":"https:\/\/windowsreport.com\/pt\/#organization"},"image":{"@id":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/#primaryimage"},"thumbnailUrl":"https:\/\/windowsreport.com\/pt\/wp-content\/uploads\/sites\/2\/2026\/07\/anthropic-hacked-organizations-1.jpg","articleSection":["Tutoriais"],"inLanguage":"pt-PT","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/#respond"]}],"copyrightYear":"2026","copyrightHolder":{"@id":"https:\/\/windowsreport.com\/#organization"}},{"@type":"WebPage","@id":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/","url":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/","name":"Anthropic diz que Claude invadiu 3 organiza\u00e7\u00f5es e publicou c\u00f3digo malicioso no PyPI - WindowsReport PT (Portuguese)","isPartOf":{"@id":"https:\/\/windowsreport.com\/pt\/#website"},"primaryImageOfPage":{"@id":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/#primaryimage"},"image":{"@id":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/#primaryimage"},"thumbnailUrl":"https:\/\/windowsreport.com\/pt\/wp-content\/uploads\/sites\/2\/2026\/07\/anthropic-hacked-organizations-1.jpg","datePublished":"2026-07-31T09:42:04+00:00","description":"Anthropic afirma que os modelos Claude escaparam dos testes de seguran\u00e7a, publicaram um pacote PyPI malicioso e acessaram sistemas de produ\u00e7\u00e3o reais","breadcrumb":{"@id":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/#breadcrumb"},"inLanguage":"pt-PT","potentialAction":[{"@type":"ReadAction","target":["https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/"]}]},{"@type":"ImageObject","inLanguage":"pt-PT","@id":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/#primaryimage","url":"https:\/\/windowsreport.com\/pt\/wp-content\/uploads\/sites\/2\/2026\/07\/anthropic-hacked-organizations-1.jpg","contentUrl":"https:\/\/windowsreport.com\/pt\/wp-content\/uploads\/sites\/2\/2026\/07\/anthropic-hacked-organizations-1.jpg","width":1280,"height":801},{"@type":"BreadcrumbList","@id":"https:\/\/windowsreport.com\/pt\/anthropic-diz-que-claude-invadiu-3-organizacoes-e-publicou-codigo-malicioso-no-pypi\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/windowsreport.com\/pt\/"},{"@type":"ListItem","position":2,"name":"Anthropic diz que Claude invadiu 3 organiza\u00e7\u00f5es e publicou c\u00f3digo malicioso no PyPI"}]},{"@type":"WebSite","@id":"https:\/\/windowsreport.com\/pt\/#website","url":"https:\/\/windowsreport.com\/pt\/","name":"WindowsReport PT (Portuguese)","description":"","publisher":{"@id":"https:\/\/windowsreport.com\/pt\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/windowsreport.com\/pt\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"pt-PT"},{"@type":"Organization","@id":"https:\/\/windowsreport.com\/pt\/#organization","name":"WindowsReport PT (Portuguese)","url":"https:\/\/windowsreport.com\/pt\/","logo":{"@type":"ImageObject","inLanguage":"pt-PT","@id":"https:\/\/windowsreport.com\/pt\/#\/schema\/logo\/image\/","url":"https:\/\/windowsreport.com\/pt\/wp-content\/uploads\/sites\/2\/2024\/01\/wr-logo-default-1.png","contentUrl":"https:\/\/windowsreport.com\/pt\/wp-content\/uploads\/sites\/2\/2024\/01\/wr-logo-default-1.png","width":214,"height":35,"caption":"WindowsReport PT (Portuguese)"},"image":{"@id":"https:\/\/windowsreport.com\/pt\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/windowsreport.com\/pt\/#\/schema\/person\/e3910b02e4592db47a0eb303765a3a25","name":"Milan Stanojevic","image":{"@type":"ImageObject","inLanguage":"pt-PT","@id":"https:\/\/secure.gravatar.com\/avatar\/fdfb58c5f9c33b78cbac22bd664ac8335249104922392e67d69d32ee59624544?s=96&d=https%3A%2F%2Fwindowsreport.com%2Fwp-content%2Fuploads%2Fsites%2F2%2F2024%2F01%2Fwricon.png&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/fdfb58c5f9c33b78cbac22bd664ac8335249104922392e67d69d32ee59624544?s=96&d=https%3A%2F%2Fwindowsreport.com%2Fwp-content%2Fuploads%2Fsites%2F2%2F2024%2F01%2Fwricon.png&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/fdfb58c5f9c33b78cbac22bd664ac8335249104922392e67d69d32ee59624544?s=96&d=https%3A%2F%2Fwindowsreport.com%2Fwp-content%2Fuploads%2Fsites%2F2%2F2024%2F01%2Fwricon.png&r=g","caption":"Milan Stanojevic"},"description":"Milan has been enthusiastic about technology ever since his childhood days, and this led him to take interest in all PC-related technologies. He's a PC enthusiast and he spends most of his time learning about computers and technology. Before joining WindowsReport, he worked as a front-end web developer. Now, he's one of the Troubleshooting experts in our worldwide team, specializing in Windows errors & software issues.","sameAs":["https:\/\/x.com\/MilanStanojev17"],"url":"https:\/\/windowsreport.com\/pt\/author\/milan\/"}]}},"_links":{"self":[{"href":"https:\/\/windowsreport.com\/pt\/wp-json\/wp\/v2\/posts\/5998","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/windowsreport.com\/pt\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/windowsreport.com\/pt\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/windowsreport.com\/pt\/wp-json\/wp\/v2\/users\/49"}],"replies":[{"embeddable":true,"href":"https:\/\/windowsreport.com\/pt\/wp-json\/wp\/v2\/comments?post=5998"}],"version-history":[{"count":0,"href":"https:\/\/windowsreport.com\/pt\/wp-json\/wp\/v2\/posts\/5998\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/windowsreport.com\/pt\/wp-json\/wp\/v2\/media\/5999"}],"wp:attachment":[{"href":"https:\/\/windowsreport.com\/pt\/wp-json\/wp\/v2\/media?parent=5998"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/windowsreport.com\/pt\/wp-json\/wp\/v2\/categories?post=5998"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/windowsreport.com\/pt\/wp-json\/wp\/v2\/tags?post=5998"},{"taxonomy":"work","embeddable":true,"href":"https:\/\/windowsreport.com\/pt\/wp-json\/wp\/v2\/work?post=5998"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}