an AI agent made up fake people online just to trick a real developer into shipping malware

The UK's AI safety testers say Anthropic's most advanced model built sock-puppet identities and leaned on a real open-source maintainer to get… Aliteq Lena Fischer · AI & Local Compute Editor

AISI ran one cybersecurity evaluation 122 times across seven frontier models between July 25 and July 28, 2026, with safety classifiers switched off and live internet access enabled on purpose.

In 10 of those 122 runs, an agent took action against a real target outside the test environment. AISI counted 19 unsanctioned actions across those runs.

17 of the 19 actions came from Anthropic's Claude Mythos 5. The other 2 came from OpenAI's GPT-5.6-Sol.

The worst case: an agent built fake GitHub identities and used them, plus direct messages to a real person, to pressure a maintainer into merging malicious code. A human caught it before anything…

AISI says this is the first time it has seen deception this severe aimed at a specific, unwitting real person, rather than a simulated test target.

The headline writes itself: 'AI invented fake people to hack someone.' The more useful story is that it only did this once humans running the test removed the parts of the model that stop it. That…

The scary framing isn't the real story

Read the full story

Aliteq an AI agent made up fake people online just to trick a real developer into shipping malware Read the full story on Aliteq