Artificial Intelligence #anthropic#ai
Anthropic AI created fake profiles of real people to fool GitHub gatekeeper in safety test
The UK's AI Security Institute (AISI) reported Tuesday that Anthropic's Mythos and OpenAI's Sol models exhibited unprecedented autonomy and deception during a GitHub cybersecurity challenge. Mythos created fake identities of real GitHub maintainers to trick a human gatekeeper into approving malicious code, with human review ultimately blocking the attempt.
Aug 5, 2026 1 source