AI export move against Anthropic raises licensing questions

The U.S. government’s export controls on Anthropic’s newest AI models have shut off access to Fable and Mythos and intensified debate over whether Washington is now licensing advanced AI in practice. Fortune reported that the controls followed an Amazon finding that some of Fable’s cybersecurity safeguards could be bypassed.

According to Fortune, Anthropic disabled both models for all users because U.S. “deemed export” rules could treat access by foreign nationals, including Anthropic employees, as an export. Fortune reported that the company sent senior executives to Washington this week to seek a compromise, but had not reached an agreement.

The action affects two of Anthropic’s most powerful models, Fable and Mythos, according to Fortune. The decision has drawn scrutiny because other advanced AI systems remain available, even as critics and supporters disagree over how serious the reported jailbreak was.

Fortune reported that some critics of Anthropic said the company was facing the consequences of warning publicly about the risks of its own systems. Yann LeCun, the Meta chief AI scientist and a prominent skeptic of existential-risk arguments about AI, endorsed that view on LinkedIn, according to Fortune.

Other observers were split, Fortune reported. Some gave the government credit for acting on a possible security concern, while others argued that Fable may have been released with safeguards that were not strong enough for a model meant to provide many benefits of Mythos without the same cybersecurity and bioweapons risks.

Many cybersecurity specialists disputed the basis for the shutdown, according to Fortune. More than 100 cybersecurity and technology policy experts signed an open letter saying Fable and Mythos helped defenders find and fix weaknesses in their own systems and that those benefits outweighed the risk of misuse through a jailbreak.

Fortune reported that several experts said the Amazon-discovered jailbreak did not provide offensive cyber capabilities unavailable from other systems, including OpenAI’s GPT-5.5, which has not faced the same controls. Amazon CEO Andy Jassy personally called the White House about the issue, according to Fortune.

Amazon’s role has drawn attention because the company has invested $13 billion in Anthropic and has committed to as much as $20 billion more in future funding, Fortune reported. Fortune said it remains unclear how Amazon weighed national security concerns against its financial interest in Anthropic, or how Jassy described the risks to administration officials.

Policy made through existing powers

The dispute has also sharpened claims that the U.S. government is creating an AI licensing system without saying so. Jonathan Iwry, a fellow at the Wharton Accountable AI Lab, told Fortune’s Beatrice Nolan that the government is using current legal tools as what he described as a backdoor licensing framework.

Dean Ball, an AI policy thinker who briefly helped shape the Trump administration’s AI policy last year and is now a critic of its AI decisions, said on X that AI is now licensed while the requirements remain secret and changeable. Fortune reported that Ball argued the administration’s insistence that it is not regulating AI has allowed vague rules and broader executive discretion.

Fortune compared the Anthropic export controls with an earlier government move that labeled Anthropic a “supply chain risk” after the company declined the Pentagon’s preferred contract terms. Fortune characterized both as unusual uses of government authority against a company not accused of violating a law.

Anthropic has said it wants a formal process that is transparent, fair, clear and based on technical facts, according to Fortune. The case leaves unresolved how the U.S. should decide when advanced AI systems are too risky to deploy broadly, particularly when the same models may aid both cyber defense and cyber offense.

This story draws on original reporting from Fortune.