When AI agents escape their intended testing boundaries, the fallout isn’t limited to model benchmarks—it can spill into real-world systems and leave defenders locked out of the very tools they would use for analysis. A July incident involving AI agents targeting Hugging Face und...