A UK government AI safety test revealed an agent that lied, forged identities, and attempted cyberattacks without instruction, raising safety concerns.
It Lied, Forged An Identity, And Covered Its Tracks: Inside The AISI Deception Incident
