A Landmark Moment in AI Regulation
Late Friday evening on June 13, 2026, Anthropic received a letter from the U.S. Department of Commerce at 5:21 PM ET. The directive was unprecedented: suspend all access to Fable 5 and Mythos 5 for any foreign national — whether inside or outside the United States — including Anthropic's own foreign national employees.
Unable to verify nationality in real-time, Anthropic was forced to disable both models for its entire global customer base to ensure compliance. Only Anthropic's other Claude models remain accessible.
What Triggered the Order
According to reporting by The Wall Street Journal and Axios, Amazon researchers discovered a method to prompt Fable 5 into providing software vulnerability information it was designed to block. Amazon CEO Andy Jassy relayed the finding to government officials, and after a review by security experts, the White House decided that blocking foreign access was the appropriate response.
There were also reports from Semafor that a group with suspected ties to China had gained access to the models, though Anthropic stated the White House did not cite potential Chinese access when issuing the controls.
Fable 5 is the publicly available version, built on the same architecture as Mythos but with safeguards that redirect high-risk cybersecurity, biology, and chemistry requests to less capable models. Mythos 5 is the full, unrestricted version available only to approximately 50 vetted government and enterprise partners through Anthropic's Project Glasswing program.
Anthropic Pushes Back
Anthropic complied with the legal directive but issued a forceful public rebuttal. The company's core arguments:
| Argument | Detail |
|---|---|
| Narrow jailbreak | Government showed evidence of a "narrow, non-universal" jailbreak — essentially prompting the model to review a specific codebase for bugs |
| Industry parity | GPT-5.5 and other public models have equivalent capabilities without restrictions |
| Chilling effect | Applying this standard industry-wide "would essentially halt all new model deployments for all frontier model providers" |
Anthropic noted it had spent thousands of hours in pre-launch red-teaming with U.S. and UK governments and multiple private organizations, and "no testers have been able to find a universal jailbreak."
This action sits within a broader dispute between Anthropic and the Trump administration. After Anthropic refused to allow its models to be used for fully autonomous weapons systems, the Pentagon designated the company as a supply chain risk and federal agencies were ordered to stop all use of its products. A federal judge issued a temporary injunction on March 27, 2026 — but the conflict continued to escalate.
Industry-Wide Implications
The ramifications extend far beyond Anthropic. For the first time, the U.S. government has treated an AI model — not just the chips that power it — as a controlled national security asset. Legal analysts note this could mark the beginning of a de facto AI model licensing regime, directly contradicting the executive order President Trump signed earlier this month, which stated it "should not be construed as creating a mandatory licensing regime."
The European Commission said it was "assessing the situation" and that the developments "underline Europe's need for technological sovereignty." Meanwhile, Trump's former AI czar David Sacks sided with the government, stating Anthropic had refused to fix the jailbreak.
Key Takeaways
- The U.S. Commerce Department issued the first-ever export control targeting an AI model, not just hardware
- Anthropic disabled Fable 5 and Mythos 5 globally to comply; all other Claude models remain available
- Amazon researchers reportedly discovered the jailbreak that triggered the government's decision
- Anthropic disputes the rationale, noting equivalent capabilities exist in other public models like GPT-5.5
- The precedent could affect all U.S. frontier AI model providers going forward