Comment by haritha1313

This just seems like lousy testing. Why was the guardrail just an understanding with the third party and a prompt and not actually tested for edge cases? Starting to wonder if its because of agents monitoring agents' work and yolo-ing it.