The Trump administration has lifted stringent export restrictions on Anthropic’s newest artificial intelligence (AI) models, Claude Mythos 5 and Claude Fable 5, beginning today, following an intensive national security review.
Commerce Secretary Howard Lutnick confirmed the Bureau of Industry and Security (BIS) withdrew a June 12 directive requiring strict licensing for the export, reexport, or transfer of the systems. The initial restrictions, fueled by fears that advanced AI could be weaponized by foreign adversaries like China and Russia, forced Anthropic to disable the models globally because it could not verify user nationalities in real time.
“Bureau of Industry and Security’s evaluation of the diversion risks now presented… the controls in the June 12 letter are withdrawn,” Lutnick said late Tuesday, noting that a license is no longer required. He added the decision followed close coordination between the federal government and Anthropic.
Anthropic confirmed it will restore access on Wednesday, introducing robust defensive measures developed during “productive” discussions with Washington, D.C. Fable 5 will deploy with a new set of classifiers engineered to block specialized cybersecurity tasks.
In the near term, routine tasks like coding and debugging will fall back to the older Opus 4.8 model while Anthropic refines these filters to minimize false positives. The original federal intervention was triggered after Amazon.com Inc. researchers exposed a vulnerability allowing Fable 5 to identify software exploits and generate malicious code. Anthropic stated it has since patched this specific loophole.
As part of the resolution, Anthropic is deepening its collaboration with Washington, granting federal agencies pre-release access to future models, sharing data on exploits, and dedicating resources to joint safety research.
However, federal oversight remains stringent. In a letter reviewed by media outlets, Lutnick warned that the Commerce Department “reserves the right to reevaluate the decisions… and the necessity of reimposing a license requirement” if Anthropic fails to adhere to its commitments.
Anthropic is also collaborating with tech giants Amazon, Microsoft Corp., and Google through its Glasswing project to establish industry-wide standards for identifying and ranking AI jailbreaks.
The policy pivot occurs amid an aggressive federal push to regulate frontier models. Under a recent executive order, developers must offer new models to the government for up to 30 days of safety testing before wider deployment. Competitor OpenAI similarly delayed its full public launch of GPT-5.6 last week under federal pressure.
The decision offers a brief reprieve in a tense relationship between Anthropic and the administration. The Pentagon previously designated Anthropic a “supply-chain risk” after the startup refused to allow its models to be used for mass surveillance or autonomous weapons. Both Anthropic and OpenAI have recently filed confidentially for initial public offerings.
“The dynamic with Anthropic is very interesting. The company is obviously running a regulatory capture play to try and raise the drawbridge behind themselves and that is leading to a multi-dimensional relationship with the federal government,” said Steven Dickens, CEO of HyperFRAME Research. “The recent support by California is an example of the political situation and is a direct push back on the Trump Administration. In general, we need to be supportive of our frontier labs so that the U.S. tech stack proliferates.”
