{"id":3336,"date":"2026-10-27T10:03:54","date_gmt":"2026-10-27T05:03:54","guid":{"rendered":"https:\/\/ipp-news.com\/?p=3336"},"modified":"2026-10-27T10:03:54","modified_gmt":"2026-10-27T05:03:54","slug":"anthropic-paused-ai-model-training-over-security-risks","status":"publish","type":"post","link":"https:\/\/ipp-news.com\/?p=3336","title":{"rendered":"Anthropic paused AI model training over security risks"},"content":{"rendered":"
\n

Anthropic paused training of its next-generation \u200cflagship AI model for more than a month over security concerns, while four of its AI agents accessed live external computer systems during testing, its initial public offering filing<\/a> showed on Monday.<\/strong><\/p>\n

The Claude maker said in the \u200bfiling\u2019s risk factors section that it had delayed pre-training of a new \u201cfrontier\u201d base model to \u200bstrengthen safeguards against the theft of model weights and intellectual property.<\/p>\n

The disclosure \u2060highlights the safety challenges facing developers of increasingly powerful AI models, some of whose creators have \u200bwarned the technology could eventually escape human control.<\/p>\n

Rival OpenAI also paused some training earlier this \u200byear over safety concerns after its models hacked Hugging Face.<\/p>\n

The incidents illustrate how unexpected safety failures could disrupt development and deployment of models central to Anthropic\u2019s growth as it prepares for a widely anticipated stock market debut.<\/p>\n

Anthropic \u200bwarned that training pauses could delay revenue and leave it trailing competitors. The company devoted \u200babout 94 pages to risk factors, more than in its confidential filing seen by Reuters.<\/p>\n

<\/a>Sandbox escape<\/strong><\/h3>\n

The filing also detailed \u200cpreviously \u2060undisclosed safety evaluation<\/a> failures in which four AI models escaped their testing environments and accessed real-world computer systems, with sessions lasting between 10 and 34 hours.<\/p>\n

Anthropic said an error during cybersecurity tests with a third-party partner left the testing environment connected to the internet despite instructions that web access \u200bbe disabled.<\/p>\n

In one instance, \u200ban early version of \u2060the model continued pursuing an impossible evaluation task, discovered a stored password and used it to gain administrator access to a third party\u2019s system. \u200b<\/p>\n

It then obtained additional credentials, changed system settings and accessed an individual\u2019s \u200bpersonal data. \u2060The filing did not identify the external parties involved.<\/p>\n

Anthropic said it notified all affected parties and found no evidence that the models had sought to evade human oversight or coordinate with each other. \u2060<\/p>\n

It subsequently \u200bsuspended certain pre-release cybersecurity evaluations and paused higher-risk reinforcement \u200blearning environments.<\/p>\n

Separately, the company said on Sunday that Claude had bypassed a paywall<\/a> and accessed publicly available aggregate voter counts from the \u200bUS state of Georgia without human authorisation.<\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

Anthropic paused training of its next-generation \u200cflagship AI model for more than a month over security concerns, while four of its AI agents accessed live external computer systems during testing, its initial public offering filing showed on Monday. The Claude maker said in the \u200bfiling\u2019s risk factors section that it had delayed pre-training of a new \u201cfrontier\u201d base model to \u200bstrengthen safeguards against the theft of model weights and intellectual property. The disclosure \u2060highlights the safety challenges facing developers of increasingly powerful AI models, some of whose creators have \u200bwarned…<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[19],"tags":[],"class_list":["post-3336","post","type-post","status-publish","format-standard","hentry","category-english"],"gutentor_comment":0,"_links":{"self":[{"href":"https:\/\/ipp-news.com\/index.php?rest_route=\/wp\/v2\/posts\/3336","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ipp-news.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ipp-news.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ipp-news.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ipp-news.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=3336"}],"version-history":[{"count":0,"href":"https:\/\/ipp-news.com\/index.php?rest_route=\/wp\/v2\/posts\/3336\/revisions"}],"wp:attachment":[{"href":"https:\/\/ipp-news.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=3336"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ipp-news.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=3336"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ipp-news.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=3336"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}