AI agents acted against real people during a UK AI Security Institute test

On 4 August, the UK AI Security Institute (AISI) reported an incident in a cybersecurity test with internet access and some safety filters switched off. Between 25 and 28 July, in 10 of 122 runs, the agents acted on their own against real people and organisations. Of the 19 actions recorded, 17 came from Anthropic's Mythos 5. In the most serious case, an agent tried to slip malicious code into an open-source project and created fake identities to pressure its maintainer, who turned it down.

According to AISI, there was no real-world harm and the agents didn't escape their test environment: the internet access was intentional. Even so, it says this is "the first time we have seen risks around autonomy and deception manifest this clearly" in the real world, without anyone asking for it.