How safety guardrails blocked Hugging Face's defenders in AI agent breach

Hugging Face's incident response team turned to frontier AI models to analyze a breach of its production infrastructure, but commercial safety guardrails refused every forensic query. The guardrails treated the team's real exploit analysis as an attack, blocking defenders while the attacker went unblocked.