New analysis of 100M forum posts finds AI tools largely ineffective for hackers, with guardrails working.
