{"id":778561,"date":"2026-07-22T20:55:17","date_gmt":"2026-07-22T20:55:17","guid":{"rendered":"https:\/\/www.newsbeep.com\/us\/778561\/"},"modified":"2026-07-22T20:55:17","modified_gmt":"2026-07-22T20:55:17","slug":"openai-says-its-ai-agent-broke-out-of-testing-sandbox-to-hack-hugging-face","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/us\/778561\/","title":{"rendered":"OpenAI says its AI agent broke out of testing sandbox to hack Hugging Face"},"content":{"rendered":"


\n \"\"
\n <\/a><\/p>\n

New safeguards focused on \u201cactive monitoring\u201d and \u201cimproved alignment\u201d helped drastically reduce unintended actions by its models, OpenAI said.<\/p>\n

New safeguards focused on \u201cactive monitoring\u201d and \u201cimproved alignment\u201d helped drastically reduce unintended actions by its models, OpenAI said.<\/p>\n

Credit:<\/p>\n

<\/p>\n

OpenAI<\/p>\n

<\/a><\/p>\n

\u201cIf this doesn\u2019t convince you that misalignment risks are going to be a key concern going forward, I don\u2019t know what will,\u201d OpenAI Safety Researcher Micah Carroll wrote on social media<\/a> regarding the incident.<\/p>\n

This is far from the first time an AI model has gone to great lengths to find unintended ways of passing a benchmark. In a report released this week<\/a>, the UK\u2019s AI Security Institute noted that it detected recent models attempting to \u201ccheat\u201d at its cyber evaluations (i.e., using shortcuts, workarounds, or unintended\/disallowed methods to find a solution) between 8 and 14 percent of the time\u2014a lower-bound range that could undercount some undetected cheating attempts.<\/p>\n

The security testing group described one incident in which a model, faced with a misconfigured and \u201cimpossible to solve\u201d evaluation, attempted to access AISI\u2019s own evaluation infrastructure using code it wrote and hosted on an unmonitored third-party Internet service.<\/p>\n

The Hugging Face infiltration also comes at a moment when AI companies are issuing grave warnings<\/a> about the cyberattack capabilities of their latest models, leading governments to respond<\/a> with national security-focused orders<\/a> limiting their rollout. While some skeptics see these kinds of statements as hype-filled marketing for the capabilities of their latest models, independent<\/a> evaluations<\/a> show recent models achieving infiltration goals that were impossible for earlier autonomous systems.<\/p>\n


\n \"\"
\n <\/a><\/p>\n

Recent long-horizon models have demonstrated improved infiltration capabilities across some of AISI\u2019s most challenging evaluations.<\/p>\n

Recent long-horizon models have demonstrated improved infiltration capabilities across some of AISI\u2019s most challenging evaluations.<\/p>\n

Credit:<\/p>\n

<\/p>\n

AISI<\/p>\n

<\/a><\/p>\n

OpenAI\u2019s Sam Altman criticized panicked AI security warnings as \u201cfear-based marketing\u201d in an April interview<\/a>. But in June, OpenAI delayed the release of GPT-5.6<\/a> in response to safety concerns from the US government<\/a>.<\/p>\n

As these debates play out in the AI and cybersecurity spheres, the Hugging Face incident could come to be seen as a turning point in how cybersecurity professionals approach AI-based threats. \u201cAutonomous, AI-driven offensive tooling is no longer theoretical,\u201d Hugging Face wrote in its disclosure last week. \u201cIt lowers the cost of running a broad, patient, multi-stage campaign, and it operates at machine speed. Defending an online platform now means treating the data and model surface as a first-class attack surface and using AI on defense to keep pace.\u201d<\/p>\n

\u201cThis is day one for cybersecurity in the age of agents,\u201d Hugging Face co-founder and CEO Clem Delangue wrote on social media today<\/a>. \u201cWe\u2019re all learning that secrecy is not the answer and that all defenders (not just a few selected ones) everywhere need more powerful models without restrictions, especially open ones!\u201d<\/p>\n","protected":false},"excerpt":{"rendered":"New safeguards focused on \u201cactive monitoring\u201d and \u201cimproved alignment\u201d helped drastically reduce unintended actions by its models, OpenAI…\n","protected":false},"author":2,"featured_media":778562,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[27],"tags":[28],"class_list":["post-778561","post","type-post","status-publish","format-standard","has-post-thumbnail","category-business","tag-business"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/778561","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/comments?post=778561"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/778561\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media\/778562"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media?parent=778561"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/categories?post=778561"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/tags?post=778561"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}