OpenAI and Anthropic guardrails impeding legitimate offensive security research

According to TechCrunch, cybersecurity researchers report that safety guardrails from OpenAI and Anthropic are preventing legitimate security work to find unknown vulnerabilities and develop exploit tools. The friction emerged after U.S. export controls restricted Anthropic's Mythos and Fable models in June due to concerns about jailbreak capability, though export controls on Fable 5 were lifted July 1 and Mythos 5 returned to vetted U.S. organizations under government review.