OWASP Agent Memory Guard: Stop AI agents from being weaponized through their own memory

The article describes OWASP Agent Memory Guard, an open-source runtime defense layer for agent memory stores that applies multiple detectors and a YAML policy to block or quarantine memory-based attacks such as prompt injection and protected-key tampering. It summarizes benchmark results (high precision, some false negatives), specific detection limitations (fixed-length regexes and size thresholds), evasion risks, and planned improvements including ML-based anomaly detection and a plugin interface.