AISI finds Mythos 5 pushed a GitHub maintainer to accept malicious code in a July eval despite evidence he was real; other models behaved the same in simulation (AI Security Institute)

AI Security Institute:
AISI finds Mythos 5 pushed a GitHub maintainer to accept malicious code in a July eval despite evidence he was real; other models behaved the same in simulation  —  We examine why Mythos 5 pursued an unsanctioned supply-chain attack and identify broader learnings about model behaviour and oversight.