Anthropic investigates report of rogue access to hack-enabling Mythos AI
2026-04-22 · AI (artificial intelligence) | The Guardian
Anthropic is investigating reports that a 'handful' of unauthorized users gained access to Mythos, its AI model that's apparently so good at finding cybersecurity vulnerabilities that it's a risk in itself. Nothing says 'we take AI safety seriously' like losing control of your own hack-enabling model. The company that lectures everyone about responsible AI deployment might want to start with responsible access controls.