Anthropic Export Controls Signal Tougher AI Regulation

The Trump administration slapped export controls on Anthropic's latest AI models last week, citing the company's failure to address a security vulnerability that allowed a "jailbreak" of the Fable 5 system. The move bans use of Fable 5 and Mythos 5 by any foreign national, inside or outside the United States, under a Commerce Department order. For the AI sector, the episode marks the first real test of the executive order on AI guardrails signed earlier this year.

A senior administration official told FOX Business that Anthropic's "recklessness" in responding to initial concerns triggered the controls. The official said the government asked the company to pull the Fable 5 model within 90 minutes after Amazon AI experts found a workaround that unlocked the model's full cyber capabilities. Anthropic had claimed those capabilities were constrained.

Anthropic pushed back. A source close to the company said the initial call from the government contained no specific details, only a demand to take down the model. The source added that CEO Dario Amodei and senior staff were engaged from the start, and that the company never refused to fix the issue. The source also denied a claim that Amodei was at a wellness retreat and unreachable, saying executives were in touch with the White House within 15 minutes.

The administration's view is different. The senior official said the experience damaged trust because Anthropic did not take the request seriously enough, given the expectations set before the Fable 5 rollout. Multiple sources said the government had made clear this release would be the first test case for the new executive order. A Washington, D.C., tech policy expert called it "government messaging 101" – if a company leads the conversation on AI safety but then appears reluctant to address safety concerns, the administration will take a tough stance.

For the broader AI sector, the implications are direct. The export controls originated at Commerce but now involve Treasury, which has expertise in computing power and markets. A former Trump administration official said the president is business-friendly only to a point, and that if a company is slow to address issues, he will force changes. The former official expressed concern that the controls will not be easily removed if Anthropic continues to brush off the issues.

Other AI developers are watching closely. OpenAI, Google, and Meta all face similar scrutiny under the executive order. The episode suggests that leading with safety rhetoric does not guarantee a softer regulatory touch. Companies that fail to engage quickly on vulnerabilities could face the same export restrictions, several policy analysts said.

Anthropic said in a June 12 statement that the vulnerabilities identified were "relatively simple" and that other publicly available models could discover them as well without requiring a bypass. The company's senior technical staff are now in Washington, D.C., meeting with White House officials. Virtual meetings have been held every day since the initial outreach, another source said.

The outcome of those talks will determine whether the export controls stick or get lifted. For now, the message to the AI industry is clear: the government is willing to use its most powerful tool – export restrictions – to enforce safety standards. The next company that faces a similar test will have to decide how fast to respond.