Commerce Secretary Howard Lutnick cleared Anthropic's most powerful AI models, Fable 5 and Mythos 5, for export after the company agreed to help write U.S. security standards. The deal does not resolve Anthropic's separate lawsuit against the Pentagon.
Commerce Secretary Howard Lutnick lifted U.S. export restrictions on Anthropic's frontier AI systems on Wednesday. Reuters and Al Jazeera reported that the Commerce Department no longer requires a license for foreign customers to use Anthropic's two most powerful models, known as Fable 5 and Mythos 5. Anthropic confirmed the change the same day.
The arrangement is conditional. Anthropic has agreed to work with the U.S. government on security standards for frontier AI and to monitor its systems for malicious use. The Commerce Department is treating the deal as a template for how future releases from any major American lab will be cleared, or held back, for export.
That template is the part of Wednesday's news that will outlast the day. Until this week, there was no public template for handling the export of general-purpose frontier AI, the kind that can write code, analyze documents, and automate tasks once reserved for knowledge workers. Earlier this year, the Trump administration ordered Anthropic to lock out foreign users from its most powerful systems, citing national-security and vulnerability concerns. The company complied by shutting off access to Fable 5 and Mythos 5 for users outside the United States.
Wednesday's decision reverses that order. The standoff, though, was never only about access. Anthropic sued the Department of Defense over concerns about military use of its AI tools, and that case is still active. The export-control question and the Pentagon lawsuit are separate legal fronts, and only one closed on Wednesday.
What is left, then, is an arrangement rather than a clean win. The Commerce Department gains an inside role in writing security standards for frontier AI. Anthropic regains the right to sell its largest models abroad. The next stress test will be the next major release from Anthropic, or from OpenAI, Google DeepMind, or Meta, where the standards drafted under this arrangement will have to decide what is safe to ship, and to whom.
The naming itself hints at how new this corner of policy is. Anthropic's public model line is best known as Claude. "Fable 5" and "Mythos 5" appear in this week's reporting as the designations for the systems the controls targeted. The fact that those names had to be reconciled across five outlets reporting the same story is a reminder that frontier-AI export controls are still being written, often before the public vocabulary catches up.
For now, the deal is the closest thing the United States has to a frontier-AI export policy. Whether it holds will depend on the next contested release, and on whether the standards it produces can keep pace with the models they are meant to govern.