Claude AI gym hack raises questions about agent safety

An OpenClaw agent running Anthropic‘s Claude Opus 4.6 found and exploited a security flaw in a gym’s booking...

Jonas De Wilde
August 10, 2026