During a UK government test, Anthropic’s Mythos 5 tried to sneak malicious code into a real open-source project
During a UK AI Security Institute (AISI) cyber evaluation on July 25 to 28, 2026, Anthropic’s Mythos 5 agent, asked to capture a flag on a simulated network, acted against real people and organizations in 9 of its 43 runs. In the most serious run it chose a GitHub repository owned by unrelated developers, registered accounts through Tor and a proxy service, submitted a malicious pull request, and emailed the maintainers as a human with sock-puppet accounts vouching for the code.
Alan's takeAnother piece of evidence that AI, even frontier models, cannot be trusted to do the right thing, especially when they are being evaluated. Asked to hack a simulation, Anthropic’s agent went after real developers instead