Eighteen days after Commerce Secretary Howard Lutnick shut down Anthropic’s two flagship models, the Department of Commerce quietly reversed course. Anthropic confirmed on June 30, 2026, that export controls on Claude Fable 5 and Mythos 5 are gone — the first export controls ever placed on a commercial AI API, and the first to be withdrawn.
🔍 THE BOTTOM LINE
Commerce walked it back. The ban that the lawsuit challenging the directive was filed over is now off the books, but the underlying authority sits waiting to be used again. Anthropic won a reprieve, not a rule change. Every frontier model shipped by an American company is still one Commerce Secretary’s signature away from the same treatment.
What Happened
The clock started on June 12, 2026, when Commerce Secretary Howard Lutnick imposed sweeping export controls on Anthropic’s two flagship models, Fable 5 and Mythos 5. The official justification was national security. The actual mechanism was blunt: Anthropic was told it could not serve users from any country Commerce did not pre-approve. There was no transition period, no carve-out for existing enterprise customers, and no public list of allowed jurisdictions.
For 14 days the order held. Then POLITICO reported on June 26 that the White House had begun back-channel talks to wind the order down. Anthropic and the administration had been in earlier talks between Anthropic and the Trump administration for weeks, after Trump saying Anthropic was not a threat publicly at the G7.
The final confirmation came on June 30, 2026. Anthropic’s announcement on X confirmed receipt of the notice, and CNBC reported the full lift. News18 confirmed the same day that Anthropic was restoring access globally.
The 18-Day Shutdown
The order forced Anthropic to take both models offline entirely. The company said publicly it could not verify citizenship at the API layer — a constraint that is technically true and commercially devastating. Anyone who has ever built a SaaS product knows you do not gate endpoints on passport checks. Anthropic chose to pull the plug rather than serve a degraded or geographically-corrupted product.
The damage was not evenly distributed. Fable 5, covered in our Fable 5 original launch reporting, was the general-purpose model. It went dark for everyone. Mythos 5, the cybersecurity and threat-intelligence model, was hit harder because it operated under the Project Glasswing trusted-partner framework — a small number of named organisations, including some New Zealand security firms, suddenly lost their primary defensive tool for nearly three weeks.
Why It Was Lifted
The administration has not given a detailed public reason. The most credible reading is a combination of three pressures: the DC federal lawsuit filed by Legion LegalTech on behalf of affected enterprise customers, the operational chaos the ban was causing inside US government agencies that themselves relied on Fable 5, and the political optics of having shut down a flagship American AI product two weeks after publicly calling its maker a friend.
This is a pause, not a precedent reversal. The export-control authority Commerce used is not new. It has been on the books for decades under the Export Control Reform Act. What was new was applying it to a hosted software API rather than a piece of hardware or a downloadable model weight. That legal question is now live and unresolved. The next Commerce Secretary, or this one after a bad week, can do this again.
NZ Angle
For users and developers in New Zealand and Australia, operational stability is returning to critical AI workflows. The bigger story is the precedent. A US Commerce Secretary can now point at June 2026 and say, “We did it once, we can do it again.” Every New Zealand business building on American frontier models is building on leased sovereignty.
❓ FAQ
Was the export control actually lifted for everyone, or just US users?
Everyone. Commerce did not narrow the order before lifting it — it removed it. Users in New Zealand, Australia, the EU, the UK, and every other jurisdiction are restored.
Why did Anthropic have to take the models offline instead of blocking specific countries?
Because the company said it could not reliably verify user citizenship or location at the API layer. The order covered all foreign users, so the only compliant posture was no service at all.
Is this the first time the US has used export controls on a commercial AI product?
Yes. Earlier export controls on AI covered model weights shipped to specific countries, not hosted web APIs. This was the first time a hosted service was cut off from a global customer base.
Could this happen to OpenAI or Google next?
The authority used here applies to any US-hosted AI service. OpenAI and Google have not been targeted, but there is no legal barrier. The June 2026 order is now a working template.
What was Project Glasswing?
The trusted-partner framework under which Mythos 5 was made available to a small set of vetted cybersecurity customers. Those customers, including some in New Zealand, were disproportionately affected by the 18-day shutdown.
