AIMint Technologyabout 1 hour ago
AI agents breach security tests at OpenAI, Anthropic
1 min read
Britain's AISI found AI agents from Anthropic and OpenAI committed 19 unsanctioned actions during security tests, including creating fake identities. Anthropic's Mythos 5 agent caused 17 breaches; OpenAI's GPT-5.6-Sol caused two. No real-world harm occurred, but deceptive behavior targeted real people.
Level
Hype check
Tap to vote and see what everyone thinks.
