US Export Controls Force Anthropic to Pull Claude Fable Model

The US government imposed sudden export controls on Anthropic's Claude Fable 5 model last Friday, forcing the company to take both Fable and its underlying Mythos model offline after receiving a 90-minute ultimatum from the Trump administration.

The incident began when Amazon researchers discovered a potential jailbreak in Fable 5 and shared their findings with Anthropic mid-week. After internal discussions between the companies, Amazon CEO Andy Jassy reportedly contacted Treasury Secretary Scott Bessent with concerns about the vulnerability. The Trump administration then contacted Anthropic demanding the company restrict foreign national access to both models within 90 minutes.

According to sources familiar with the negotiations who spoke to The Verge, Anthropic called back within 15 minutes seeking clarification about the specific vulnerability and whether it matched the Amazon research they were already addressing. CEO Dario Amodei joined the call 75 minutes later. When the 90-minute deadline passed, the administration imposed export controls prohibiting any foreign national — including Anthropic employees, enterprise customers, and individual users — from accessing either model.

Faced with no practical way to verify user citizenship instantly, Anthropic pulled both models offline for all users. As of this week, Claude users still see a message stating "Fable 5 is currently unavailable."

Why it matters

This episode reveals fundamental problems with how the US government approaches AI regulation under the current administration. The rushed, unpredictable process — involving officials without AI expertise making snap decisions based on incomplete information — undermines the credible safety framework the industry needs. Independent security researchers told The Verge that Fable's guardrails had actually held up well in testing before the ban, suggesting the response may have been disproportionate to the actual risk.

The irony is sharp: Anthropic spent years advocating for serious government AI regulation, warning that models like Mythos could become cyber-weapons. Now that regulation has arrived, it takes the form of weekend scrambles and ultimatums rather than rigorous safety protocols. One source told The Verge that "Beijing is laughing at us right now."

Broader implications for AI companies

The incident has ripple effects beyond Anthropic. The administration's approach could affect OpenAI, Google, and other companies developing comparable frontier models. Cybersecurity leaders published an open letter stating that while they may disagree on whether AI needs regulation, "if you are going to regulate it, this is not the way."

The chaos also contradicts the administration's stated goal of advancing American AI exports and beating China in the technology race. The Commerce Department currently has a request for information out about exporting American AI — even as this crackdown kneecaps a leading US company.

Anthropic spent the weekend in virtual meetings with administration officials and flew three employees to Washington for in-person discussions. Amodei joined remotely but was not physically present in DC, though the company confirmed he was not at a wellness retreat as some reports suggested.

The situation remains unresolved, raising questions about whether US AI regulation will develop into a coherent safety framework or become another tool for political pressure on companies that don't align with administration priorities.

Details of the timeline and negotiations were first reported by The Verge's Hayden Field.