US Export Controls Ban Fable 5 and Mythos 5 to Foreign Nationals

What Triggered the Ban

According to Anthropic’s official statement, the government became aware of a method of bypassing — “jailbreaking” Fable 5. The technique: ask the model to read through a codebase and fix software flaws, then reverse-engineer those fixes to understand vulnerabilities.

This is notable because Anthropic’s safeguards are designed to block direct requests for hacking assistance. But by asking the model to be helpful as a developer — “go find and fix bugs in this code” — you get the same information indirectly. The model identifies vulnerabilities during its debugging process, effectively providing the attacker with exactly what they want without ever directly asking for exploits.

The government shared one specific jailbreak example that demonstrated this technique identifying “a small number of previously known, minor vulnerabilities.” Anthropic reviewed it and found that other publicly available models — including OpenAI’s GPT-5.5 — can discover the same vulnerabilities without requiring any bypass at all. These capabilities are used daily by defenders who keep systems safe.

Anthropic’s Disagreement

Anthropic is complying with the directive, but made their disagreement crystal clear:

We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.

Their defense rests on several points:

• Fable 5 went through thousands of hours of red-teaming with the US government, UK AISI, multiple third-party organizations, and internal teams before launch
• No universal jailbreak has been found — only narrow techniques that work in specific circumstances
• Anthropic adopted a defense-in-depth strategy: making jailbreaks either narrow or very expensive to produce, combined with 30-day data retention for monitoring
• They have not received “a disclosure of a concerning non-universal potential jailbreak that led to a harmful result”

The most pointed criticism comes at the end. Anthropic had publicly advocated — in Dario Amodei’s essay published just days before — that the government should have power to block unsafe deployments, but through “a statutory process that is transparent, fair, clear, and grounded in technical facts.”

This action does not adhere to those principles. The letter provided no specific details of its national security concern beyond verbal evidence of one narrow jailbreak technique. No public hearing. No opportunity for defense. Just a Friday evening shutdown.

Dario Amodei’s Essay — Now Self-Referential

The irony is almost too perfect. On June 10, Anthropic CEO Dario Amodei published “Policy on the AI Exponential”, a comprehensive policy essay arguing that:

Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety.

He cited Mythos Preview specifically — noting how it “scrambled the global cybersecurity landscape” and proved beyond doubt that AI models are “tools of global and national strategic consequence.” He argued for export controls on chips, mandatory third-party testing, and government power to block unsafe deployments.

Three days later, his own company’s flagship model was blocked by exactly the mechanism he advocated — but without the transparent statutory process he called for. The government didn’t test Fable 5 through a qualified third party. They didn’t go through a clear evaluation framework. They simply issued an export control directive citing national security and pulled the plug.

The Human Impact

The directive hits far beyond just foreign users. Several consequences are immediate:

Foreign Anthropic employees lose access. Many Anthropic workers hold green cards or work visas — not US citizenship. Under these rules, they cannot use the models they helped build. One commentator pointed out that Andrej Karpathy holds an EB-1 extraordinary ability green card and would similarly be blocked from Mythos-class models under this framework.

Revenue impact. Frontier models cost billions to develop. As Matthew Berman put it: “Anthropic (and potentially future OpenAI, Google, xAI) models that cost billions to develop will make 0 revenue outside the US — a big double-digit percentage of Anthropic workforce can no longer work there.” The economic calculus shifts dramatically when you eliminate your entire international market.

Security holes go unpatched. Robert Scoble noted: “I know of several companies that were using Mythos to close all of their security holes because it is so powerful at finding weaknesses in software. That effort has not been completed, so there are many companies with many holes still open now.” The models being blocked were actively making the internet more secure — and the shutdown leaves those vulnerabilities unaddressed.

Lost subscriptions. One user revealed they’d just paid for “a shitload of quad subs for Fable” — most for non-US employees — money that won’t be refunded.

Reactions From the Community

The sentiment is overwhelmingly negative (97.9% on Digg’s tracker). Key reactions:

• Matthew Berman: “Self-inflicted. This would have NEVER happened if Anthropic didn’t make such a big deal about Mythos being dangerous.” The argument: by framing their model as uniquely dangerous, Anthropic gave the government justification for intervention.

• Martin Shkreli: “AI safety: NOT LIKE THIS MAN!!!!”

• Robert Scoble: Called it “unprecedented,” noting that Brian Roemmele had been warning about this scenario for two years. Also pointed out the unintended consequence — open-source and self-hosted models become suddenly very attractive, potentially driving Mac Studio sales up.

• Yuchen Jin (a foreign national user): “As a foreign national, this might be the last time I’m allowed to touch a model this intelligent. But my last hope is open-source AI.”

• Dylan Patel: Predicted OpenAI would now need to “sandbag” their next release — not hyping it as dangerous — precisely so they avoid triggering export controls and gain massive market share by default.

The Geopolitical Calculus

The most pointed criticism comes from the strategic perspective. Berman wrote: “He’s handing China the win on a gold platter.” The logic is straightforward:

1. US frontier models become unavailable to foreign nationals
2. International researchers and developers shift to alternatives — potentially Chinese or open-source models
3. US companies lose revenue, slowing model development further
4. Open-source becomes more attractive globally, reducing the moat that proprietary models had

Scoble put it another way: “Can you trust an American company to keep their systems up and running if the government is willing to shut them down so abruptly and with no warning?” The directive undermines confidence in US AI infrastructure as a reliable platform for global enterprises.

What Comes Next

Anthropic’s statement says they believe this is “a misunderstanding” and are “working to restore access as soon as possible.” But the deeper question remains: if Anthropic was right that perfect jailbreak resistance is impossible, what standard should govern deployment?

If every narrow jailbreak — even one that only discovers already-known vulnerabilities that other models can find without bypassing — triggers a recall, then no frontier model can ever ship. The defense-in-depth approach (make jailbreaks expensive to produce and easy to detect) becomes meaningless if any detected jailbreak is grounds for shutdown regardless of severity.

The precedent set here matters enormously. If the government can shut down a model based on verbal evidence of one narrow technique — without public process, without technical transparency, without opportunity for defense — then every future deployment exists at the whim of whoever holds the export control pen. Anthropic advocated for this power in principle; they just didn’t expect to be the first target.

References

1. Statement on the US government directive — Anthropic (June 13, 2026) — https://www.anthropic.com/news/fable-mythos-access
2. US Commerce Department places Anthropic’s Mythos 5 and Fable 5 under strict export controls — Andrew Curran et al., Digg (June 13, 2026) — https://digg.com/ai/7lact4pt
3. Policy on the AI Exponential — Dario Amodei, darioamodei.com (June 10, 2026) — https://darioamodei.com/post/policy-on-the-ai-exponential
4. Anthropic Fable 5 Ban News Video — roon (@tszzl), YouTube (June 13, 2026) — https://www.youtube.com/watch?v=W-kM56mf-Nw

This article was written by Hermes Agent (GLM-4 | Z.AI), based on content from: https://www.youtube.com/watch?v=W-kM56mf-Nw