AI security

2 posts with this tag

BenchJack: Scans AI Agent Benchmarks for Hackability Vulnerabilities

BenchJack audits AI agent benchmarks to detect hackability flaws like leaked keys, unsafe evaluations, and prompt injections that let models cheat without real capability. Designed for developers and researchers, it employs static tools including Semgrep and Bandit plus AI-driven analysis with Claud

Administrator 5/2/2026

SlowMist Agent Security: Stop AI Agent Exploits Before They Happen

AI agents face real threats: prompt injections, poisoned RAG data, and malicious tool calls. This deep dive explores SlowMist Agent Security — a 290-star open-source toolkit designed to detect and prevent critical agent exploits before they compromise your systems.

Administrator 4/1/2026