The Trump administration has escalated its stance on frontier AI, blocking foreign governments, companies and individuals from accessing Anthropic's most advanced models. Commerce Secretary Howard Lutnick sent a letter Friday to CEO Dario Amodei declaring that the Mythos 5 and Fable 5 systems would be subject to export controls—both for any location outside U.S. borders and for all foreign persons operating within them.

The Export Control Framework

Per Commerce's directive, Anthropic now requires a license for the export, re-export, or domestic transfer of its flagship models. That's not all—the company must also submit additional applications for individually validated licenses to operate at scale. Non-compliance carries financial and civil penalties. To top it off, Anthropic has landed on a Pentagon blacklist deeming its systems too dangerous for government use.

The Jailbreak That Triggered This

An administration official told Axios the decision came after another company claimed it successfully jailbroke Mythos 5, raising alarms about potential national security implications. Sources suggest the administration attempted to convince Anthropic to voluntarily pause releasing the latest models before shipping—talks that apparently went nowhere. The official indicated the models need to remain locked down until federal security infrastructure catches up, estimating that hardening could take "the next few weeks."

Voluntary Testing vs. Hard Controls

The export controls arrive just weeks after an executive order requiring pre-deployment testing for advanced AI models—though notably, that framework is voluntary and explicitly avoids a licensing regime. White House chief AI adviser David Sacks reportedly pushed hard to keep it that way, avoiding what he characterized as "regulatory capture" of the major labs. An administration official emphasized Trump "does not want to hurt the industry and wants innovation to continue," suggesting these moves aren't meant to stifle development but to gatekeep access.

Key Takeaways

  • Mythos 5 and Fable 5 are now under export controls for all foreign persons, including those inside the U.S.
  • Anthropic faces a Pentagon blacklist that bars government agencies from using its models directly.
  • A third-party jailbreak claim reportedly prompted the action after voluntary release delays failed.
  • Pre-deployment testing remains voluntary despite new hard controls on model distribution.

The Bottom Line

This is Washington drawing a line in the sand on frontier AI—treating it like dual-use infrastructure rather than software. Anthropic's caught between federal pressure to lock things down and its own push to ship. That tension isn't going away anytime soon.