Why it matters
The AgentFuzz researchers present directed greybox fuzzing for finding paths from attacker-controlled natural-language input to security-sensitive agent operations. Their evaluation across 20 open-source agents combines generated seed prompts, semantic and distance feedback, and argument-aware mutation, reporting 34 high-risk zero-days and 23 assigned CVEs.
My takeaway: Combine prompt mutation with program-level reachability signals instead of judging only model output. Instrument tool arguments and sensitive sinks, prioritize seeds that move execution toward those sinks, reproduce every finding without the model in the loop, and retain confirmed paths as deterministic regression cases.