The Trump administration has begun softening its restrictions on Anthropic’s powerful cybersecurity AI models, allowing Mythos 5 to be made available to more than 100 specific US government agencies and companies. The directive, issued by Commerce Secretary Howard Lutnick to Anthropic’s chief compute officer Tom Brown, confirmed that appropriate safeguards were in place to permit certain trusted partners to access the model, including non-American employees at those organisations — a group explicitly caught by the original ban.
Anthropic publicly acknowledged the development, stating it had been working closely with the US government since June 12 to restore access, and that Mythos 5 would be redeployed quickly to organisations operating and defending critical infrastructure. The company said it was continuing to work toward expanded access for Mythos 5 and a full return of Fable 5 to general availability. The administration’s directive did not address Fable 5, the more widely released version of Mythos that was pulled alongside its more powerful counterpart after security researchers reportedly bypassed its guardrails.
The partial restoration comes as the two-week ban has already begun reshaping the competitive landscape in Asia. Japanese AI startup Sakana AI, co-founded by Google alumni David Ha and Llion Jones alongside former Mercari and Stability AI executive Ren Ito, launched Fugu, a frontier model it describes as comparable to Fable 5 and Mythos Preview, explicitly marketed as delivering frontier AI capability without exposure to US export controls.
Simultaneously, Chinese cybersecurity firm 360, founded by Zhou Hongyi, unveiled two AI security tools: Tulongfeng, designed to automatically discover software vulnerabilities, and Yitianzhen, built for automated cyber defence and incident response. Hongyi framed vulnerability-finding AI as a national strategic asset, warning against scenarios where advanced capabilities remain accessible to only some actors.
With Anthropic’s run-rate revenue reported at $47 billion annually as of May 2026, the extent of any commercial damage from the ban remains unclear — but local AI alternatives in both Tokyo and Beijing have already moved to occupy the space it left behind.
