Safety guardrails hindered Hugging Face's incident response to AI breach

Frontier AI models blocked forensic queries from Hugging Face's incident response team during a production infrastructure breach. The safety guardrails incorrectly identified the team's exploit analysis as malicious, preventing them from using the models to investigate the attack.