News · 2026-06-30

The US fully lifts its export ban on Anthropic's most powerful models

The US government fully repealed export controls on Anthropic's Fable 5 and Mythos 5 models on June 30, just eighteen days after imposing them. The Commerce Department placed the restrictions on June 12, partially eased them in late June, and then lifted them entirely — one of the fastest reversals of a major tech-policy decision in memory.

Key facts

• What: Two and a half weeks after restricting Fable 5 and Mythos 5, Washington reversed course completely, ending the licensing requirement to send the models abroad.
• When: 2026-06-30
• Primary source: read the source

On June 12, the Commerce Department placed export controls on two of Anthropic's top models, Fable 5 and Mythos 5. Export controls restrict what advanced technology can be sold or sent outside the country without a special license — the same toolkit long used for weapons components and, more recently, for the high-end chips that train AI. Applying them to the AI models themselves, not just the hardware, was the striking part, as we covered in the US government banned Anthropic's most powerful AI model. Around the end of the month the controls were partially eased, which we noted in the Anthropic model ban partially lifts. On June 30, the controls were removed completely. Reuters reported the full reversal, and Anthropic confirmed it in a post on X, saying access would begin to be restored the following day. Fox Business reported the same, tying the decision to Anthropic's work with the government.

According to Commerce Secretary Howard Lutnick, the department ran a national-security review focused on diversion risks — the worry that a restricted model could end up, through resale or rerouting, somewhere the US did not want it. After that review, and after what officials described as cooperation from Anthropic, they concluded the restrictions were doing more to hobble an American company than to protect anyone, and framed lifting them as a way to support America's leadership in AI. In plain terms: the government decided the leash was hurting its own runner.

Controlling a model is fundamentally different from controlling a chip. A chip is a physical object; you can inspect it at a border, count them, track a shipment. A model is a large file of numbers — the trained weights — that can be copied perfectly and moved anywhere in seconds. Trying to export-control software that can be duplicated at will is like trying to control the spread of a recipe by regulating flour. That mismatch is a big reason the original restriction drew doubts from the start, and part of why unwinding it was straightforward once the political will shifted.

The episode matters beyond one company's product line. For the first time, a government treated a specific commercial AI model the way it treats sensitive hardware — and then discovered how awkward that fit is in practice. The reversal does not settle the underlying question of whether, and how, frontier models should be governed at the border; it mostly shows that the tools built for a world of physical goods do not map cleanly onto files that copy for free. Expect more improvisation as other governments test the same idea.

A lot here rests on official statements and a source-based news report rather than a published rule with its full reasoning laid out, and the speed of the turnaround invites the question of how carefully considered the original controls were in the first place. If a restriction billed as a national-security measure can be imposed and fully rescinded inside three weeks, it is fair to ask whether it was ever load-bearing, or whether it was a negotiating posture that resolved once both sides talked. For companies building on frontier models, the practical takeaway is less about this one decision and more about the volatility around it: the ground rules for who can use the most capable models, and where, are still being written in real time — and they can move faster than any product roadmap. We track the wider fallout in the model ban is quietly redrawing the AI map.

Primary source, verified: read the paper →

Topics: policy · export-controls · Anthropic · government · industry