Safety3h ago
RIFT-Bench Automates Red-Teaming For AI Agents
A new graph representation methodology called RIFT-Bench enables dynamic red-teaming across diverse agentic architectures. The system uses automated discovery and scanning phases to deploy adaptive adversarial attacks. It moves beyond static LLM vulnerability tests to target autonomous decision-making flaws. This provides a unified framework for practitioners to evaluate security across heterogeneous AI systems.