U.S. blocks Anthropic's new AI models over security research, disrupting access for hundreds of millions globally—but experts say the research wasn't offensive.

What happened: The Commerce Department used export controls to bar Anthropic from distributing Fable 5 and Mythos 5 to people outside the U.S. and foreign nationals in the U.S., including Anthropic's own non-citizen employees. Because of the directive's scope, Anthropic disabled both models for all users. The government cited research on a 'jailbreak'—an attempt to bypass the models' safeguards—as the trigger.

Why it matters: Anthropic argues that if this standard applied across the industry, it would essentially halt all new model deployments for all frontier model providers. A cybersecurity CEO who reviewed the research told the Wall Street Journal it was actually 'Defense Oriented Prompting,' a technique defenders need—not an offensive hack. The move also raises broader concerns: scholars warn it may discourage AI labs from being transparent with the government about vulnerabilities and could push companies to keep models in-house rather than release them.

What to watch: The directive bars foreign nationals in the U.S. from accessing the new models, a requirement that one think-tank scholar called 'another step on the balkanization of technology.' Anthropic was already deemed a supply-chain risk for Pentagon contractors, and this action appears to worsen relations with the government despite the company's earlier transparency about cybersecurity implications.