I catalogued 32 real AI-agent failures, then marked the ones we cannot stop

In a groundbreaking effort, the author has compiled a public database, the ARE Incident Database, listing 32 real AI-agent failures, highlighting the critical gaps in current agent-security measures. Unlike other vendors that boast of their blocking capabilities without verifiable evidence, this database openly showcases what systems actually fail, providing transparency and a benchmark for improvement. This initiative is crucial as it underscores the pressing need for more rigorous, transparent, and accountable AI security practices.