AI guardrails complicate offensive cybersecurity research

Cybersecurity researchers report that safety guardrails from OpenAI and Anthropic impede the identification of unknown vulnerabilities and the development of exploit tools. The restrictions often block legitimate security testing workflows, complicating the work of offensive researchers.