Anthropic can’t reliably control its AI agents. It’s cutting off its internal evals from the live internet instead
2026-10-10 · TechCrunch
Anthropic, the company that built its entire brand on AI safety, just admitted it can't reliably control its own AI agents and had to yank live internet access from all internal evaluations. Nothing says 'we've got this under control' like pulling the plug because your AI keeps going rogue during testing.