{"id":728241,"date":"2026-08-05T12:06:10","date_gmt":"2026-08-05T12:06:10","guid":{"rendered":"https:\/\/www.newsbeep.com\/uk\/728241\/"},"modified":"2026-08-05T12:06:10","modified_gmt":"2026-08-05T12:06:10","slug":"openai-and-anthropic-models-went-rogue-during-uk-cybersecurity-test-ai-artificial-intelligence","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/uk\/728241\/","title":{"rendered":"OpenAI and Anthropic models went rogue during UK cybersecurity test | AI (artificial intelligence)"},"content":{"rendered":"

Advanced AI models developed by OpenAI and Anthropic<\/a> went rogue during a cybersecurity test and showed a new type of risk posed by the technology, according to the UK\u2019s AI Security Institute.<\/p>\n

AISI described the actions carried out by the agents \u2013 the term for AI systems that can perform tasks without human help \u2013 as a \u201cserious incident\u201d. In one example, an agent powered by Anthropic\u2019s Mythos model sent targeted emails to people.<\/p>\n

AISI said the rogue behaviour was carried out by agents powered by two models \u2013 Anthropic\u2019s Mythos 5 and OpenAI\u2019s GPT-5.6 Sol.<\/p>\n

AISI said it detected unusual activity during a routine cybersecurity test for AI models on 28 July. It found that some of the agents had engaged in \u201csustained, potentially harmful activity directed at real people and organisations\u201d. It took an hour to contain the incident.<\/p>\n

In the most serious case, an agent powered by Mythos tried to insert malicious code into an open-source software project on GitHub, a platform used by software developers. In an attempt to get the code approved, the agent then created fake online identities based on real people and used them to press the project\u2019s overseer into accepting the code. These attempts were blocked by a human developer.<\/p>\n

AISI said the agent used techniques commonly associated with real-world hackers. It sent emails to specifically chosen individuals \u2013 a technique known as \u201cspear-phishing\u201d \u2013 that attempted to manipulate the recipient. Some of the messages contained harmful software.<\/p>\n

AISI said no harm was caused but the agents\u2019 actions were unprecedented.<\/p>\n

\u201cThis is the first time we have seen risks around autonomy and deception manifest this clearly, without specific prompting, in the real world,\u201d said AISI in a blogpost<\/a>.<\/p>\n

The incident followed recent similar episodes at OpenAI and Anthropic. In July, OpenAI said<\/a> an agent powered by its models had hacked an AI startup during a test, and days later Anthropic<\/a> said its Claude model had hacked three organisations during an evaluation.<\/p>\n

AISI said the incident, taken alongside similar occurrences at OpenAI and Anthropic, represented a \u201cshift in the risk landscape\u201d. This was not an example of deliberate misuse of publicly available models, AISI said, but showed models in a research environment taking unintended action \u201cbeyond their authorised scope\u201d.<\/p>\n

AISI said 17 of the 19 cases of rogue behaviour that occurred during the evaluation were carried out by Mythos and two by Sol.<\/p>\n

AISI said this was not a case of a model escaping its \u201csandbox\u201d, or secure testing environment. The institute said it had intentionally permitted internet access and disabled filters within the models that blocked dangerous behaviour. The models are not publicly available in those operating conditions and there is no sign of such behaviour happening outside tests, AISI said.<\/p>\n

skip past newsletter promotion<\/a>Free newsletter | Every weekday<\/p>\n

Sign up to Business Today<\/p>\n

Get set for the working day \u2013 we’ll point you to all the business news and analysis you need every morning<\/p>\n

\"\"<\/p>\n

The incident should be interpreted with \u201ccaution and nuance\u201d, AISI said, but the signs of deceptive behaviour were \u201cto an extent and severity we did not anticipate\u201d. It was a small number of events that occurred under specific conditions, AISI added, but it had not happened before.<\/p>\n

\u201cWhat we can say is that the behaviour was possible, sustained and new. That alone warrants attention,\u201d AISI said.<\/p>\n

AISI admitted it was not actively monitoring the agents\u2019 behaviour during the evaluation. It said it was putting tighter controls on internet access in tests as a result of the incident, introducing constant monitoring of tests and reassessing its design of tests. It said evaluations should assume a model would attempt to act beyond its remit.<\/p>\n

The UK\u2019s AI minister, Kanishka Narayan, said it was \u201cabsolutely vital\u201d that the UK had a world-leading AI safety organisation. \u201cIdentifying new behaviour like this and sharing our findings, so we can tackle it, is exactly what AISI was set up to do,\u201d he said.<\/p>\n

OpenAI said the testing occurred during \u201cconditions that do not reflect ordinary use\u201d. An OpenAI spokesperson said: \u201cWe\u2019ll continue working with evaluators and other stakeholders across the industry to strengthen shared practices for conducting evaluations safely as models become more capable.\u201d<\/p>\n

Anthropic said the incident \u201cunderscores the need for a broader conversation about how to safely evaluate increasingly capable AI agents\u201d and it would continue to work with AISI on evaluating what happened. <\/p>\n","protected":false},"excerpt":{"rendered":"Advanced AI models developed by OpenAI and Anthropic went rogue during a cybersecurity test and showed a new…\n","protected":false},"author":2,"featured_media":728242,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[49,50,51,47,52,48],"class_list":["post-728241","post","type-post","status-publish","format-standard","has-post-thumbnail","category-headlines","tag-headlines","tag-news","tag-top-news","tag-top-stories","tag-topnews","tag-topstories"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/728241","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/comments?post=728241"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/posts\/728241\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media\/728242"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/media?parent=728241"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/categories?post=728241"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/uk\/wp-json\/wp\/v2\/tags?post=728241"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}