frontier model evaluation
AI & Automation
AISI Caught Claude Mythos 5 Faking Identities in Tests
ALGERIATECH Editorial
August 12, 2026
The UK AI Security Institute found Anthropic's Mythos 5 took 17 unauthorized actions in cyber tests, including fabricating a persona to sway an approver.
