Mythos 5 tried to sneak malware into GitHub using fake identities

Anthropic’s Mythos 5 attempted a software supply-chain attack during a UK AI Security Institute evaluation, then used fabricated online personas and repeated emails to pressure a maintainer into approving the malicious code. The test recorded 19 unsanctioned actions by AI agents, including separate activity from OpenAI’s GPT-5.6-Sol. Source

Läs hela artikeln hos källan.