- On June 12, 2026, the US government issued an export control directive forcing Anthropic to disable Fable 5 and Mythos 5 for every customer worldwide.
- The order bars access by any foreign national, inside or outside the US, including Anthropic’s own foreign-national employees.
- Anthropic received the letter at 5:21pm ET; the stated concern is an alleged method of “jailbreaking” Fable 5.
- All other Anthropic models, including Opus, Sonnet, and Haiku, are unaffected and remain available.
What does it take for a government to switch off an AI model used by hundreds of millions of people? On June 12, 2026, the answer turned out to be a single letter that arrived at 5:21pm Eastern Time. Citing national security authorities, the US government ordered Anthropic to suspend all access to its two newest frontier models, Fable 5 and Mythos 5, and the company complied within hours. It is one of the most consequential government interventions in the commercial AI era so far, and the reasoning behind it is now at the center of a fierce dispute.
What the Directive Actually Says
According to Anthropic’s public statement, the export control directive suspends all access to Fable 5 and Mythos 5 “by any foreign national, whether inside or outside the United States,” a scope so broad that it even covers the company’s own foreign-national employees. Because Anthropic could not selectively block only foreign users without disrupting its infrastructure, the practical effect was to disable both models for the entire customer base. The company was careful to note one boundary: access to all of its other models is not affected.
A Letter With Few Specifics
Anthropic says the directive arrived at 5:21pm ET and “did not provide specific details of its national security concern.” The company’s understanding is that the government believes it has identified a way to bypass, or “jailbreak,” Fable 5. Anthropic reviewed a demonstration of the technique and concluded that it surfaced only a small number of previously known, minor vulnerabilities, ones it says other publicly available models can also discover without any bypass at all.
Trend Insight — The breadth of the order is the real story. By targeting “any foreign national” rather than specific countries or entities, the directive treats a commercial model less like software and more like a controlled dual-use technology, a framing that could become a template for future AI regulation.
Anthropic’s Pushback: A Narrow Jailbreak
Anthropic is complying with the order but openly disagrees with it. The company points to the safeguards described in its Fable 5 launch post, arguing they are “substantially more effective than those of any previously deployed model,” to the point that many users have complained they are overly broad. In the weeks before launch, Anthropic says it worked with the US government, the UK AI Safety Institute, multiple third-party organizations, and internal teams to red-team Fable’s safeguards for thousands of hours.
“Defense in Depth,” Not Perfection
Crucially, Anthropic concedes that no provider can guarantee perfect jailbreak resistance. It says no testers have found a “universal jailbreak” capable of broadly unblocking cyber capabilities, and that its strategy is “defense in depth”: make jailbreaks either narrow or very expensive to produce, then pair that with monitoring to catch attacks quickly. The 30-day data retention requirement for Fable, which the company admits carries real costs with customers, exists specifically to support that monitoring.
As for the specific technique behind the directive, Anthropic describes it as a “narrow, non-universal jailbreak” that “essentially consists of asking the model to read a specific codebase and fix any software flaws.” The company says it validated that the level of capability shown is “widely available from other models (including OpenAI’s GPT-5.5)” and is used every day by defenders who keep systems safe.
Trend Insight — The disagreement is not about whether jailbreaks exist, but about thresholds. If a narrow, replicable-elsewhere vulnerability is enough to recall a flagship model, the bar for deployment shifts dramatically, and the entire frontier sector inherits that uncertainty.
Why This Sets a Precedent for the Whole Industry
Anthropic’s sharpest argument is about precedent. It contends that recalling a commercial model deployed to hundreds of millions of people over “the finding of a narrow potential jailbreak” would, if applied as a standard across the industry, “essentially halt all new model deployments for all frontier model providers.” Because comparable capabilities exist in other widely available models, a consistent application of this logic would not stop at one company.
A Call for Due Process
The company reiterated its stated policy position that the government should be able to block unsafe deployments, but only through “a statutory process that is transparent, fair, clear, and grounded in technical facts.” It argues this action did not meet those principles, and that it has not received a disclosure of a concerning jailbreak that led to a genuinely harmful result. Anthropic apologized to customers for the disruption, called the episode a “misunderstanding,” and said it is working to restore access as soon as possible while promising more details within 24 hours.
Trend Insight — For business leaders, the lesson is about continuity risk. A model that is fully available one afternoon can be gone by evening, on grounds the vendor itself cannot fully see. Multi-model strategies and clear fallback plans are no longer optional for teams building on frontier AI.
Related
- The Largest IPO Ever Is Secretly an AI Bet
- Why Meta’s Next AI Hub Isn’t in America
- Google’s New AI Doesn’t Write Text, It Erases It
- AI Tech Digest: The Week’s Biggest Moves
Sources
- Anthropic — Statement on the US government directive to suspend access to Fable 5 and Mythos 5
- CNBC — Anthropic disables access to Fable 5 and Mythos 5 to comply with government directive
- Fortune — Anthropic disables Fable and Mythos after US bars foreign access
AI Biz Insider · AI Trends EN · aibizinsider.com
댓글 남기기