Agents2h ago
RIFT-Bench Automates Red-Teaming For AI Agents
A new framework called RIFT-Bench uses graph representation to automate adversarial attacks on autonomous AI systems. It employs a two-phase discovery and scanning process to identify vulnerabilities across diverse agent architectures. This methodology allows developers to unify security evaluations for heterogeneous LLM agents. It replaces static, domain-specific testing with adaptive red-teaming.