How Agent Anomaly Detection scores on benchmarks we didn't write

Agent Anomaly Detection (AAD) provides asynchronous, low-latency monitoring for AI agents by analyzing OpenTelemetry traces and logs to flag goal hijacks and tool misuse. Tested across five public security benchmarks, AAD successfully caught 87% of harmful instructions carried out by agents without adding runtime latency. Developers can integrate... Weiterlesen: How Agent Anomaly Detection scores on benchmarks we didn't write