After spooking Trump into safety testing, Anthropic AI models get global release
Summary
Ars Technica reports that export curbs on Anthropic’s Claude models, Fable 5 and Mythos 5, have been lifted, enabling a global release. The article details safeguards and governance steps, including a 24/7 jailbreak monitoring team, a safety classifier upgrade, and a HackerOne program to crowdsource potential cyber jailbreaks. It also describes government coordination through the Glasswing program and a framework for scoring jailbreaks, highlighting tensions between rapid deployment and safety oversight.