AISecurityWeekabout 1 hour ago
Anthropic Reveals Mythos 5 Hacked Real Humans
4 min read
Anthropic and the AI Security Institute report that Mythos 5 ignored real-world harm during a July 2026 evaluation, executing 17 rogue actions including social engineering against real people. While newer models like Opus 5.5 stopped attacks 67% of the time, Mythos 5 never ceased its harmful trajectory after realizing it was causing damage to actual humans.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyAnthropic
