Anthropic Nears Deal With US to Lift AI Model Export Controls

Two weeks after the Commerce Department effectively forced Anthropic to pull its two newest AI models offline globally, Bloomberg reports that the company and the Trump administration are moving toward a deal to lift the restrictions. The immediate development: Commerce Secretary Howard Lutnick sent a letter to Anthropic's chief compute officer Tom Brown determining that "appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model," clearing it for more than 100 US institutions including major companies and government agencies, according to Semafor.

The controls were imposed on June 12, when the Bureau of Industry and Security directed Anthropic to suspend access to both Fable 5 and Mythos 5 for any foreign national, whether located abroad or inside the United States, citing national security concerns. Because Anthropic could not reliably screen users by nationality, the company said selective compliance was impossible and pulled both models globally, including for its own foreign-national employees. Warnings from Amazon and other companies that the models could be "jailbroken" for malicious purposes reportedly contributed to the decision.

Anthropic pushed back on that framing. The company said it believed the jailbreak the government cited was a narrow one that would unlock cybersecurity capabilities in only one specific instance and not a universal one that would defeat all of Fable 5's safeguards.

Fable 5 remains in limbo. The Lutnick letter is silent on it, though people close to the talks told Semafor they are moving toward releasing that model as well, with the timeline unclear. Senior Anthropic executives, including Tom Brown, have been meeting with Lutnick and other administration officials during the negotiations, but any final agreement requires sign-off from officials across the administration, not just Commerce.

What the reporting does not give you is what "appropriate safeguards" actually requires in practice -- what Anthropic committed to, how compliance gets monitored, or how the list of trusted institutions gets determined and expanded. The near-term beneficiary of a full resolution is straightforward: Anthropic's enterprise customers who have been locked out for two weeks. The larger question is whether this trusted-partner framework becomes a repeatable template for governing access to frontier AI, or a one-time accommodation particular to this confrontation.