US Government Shuts Down Anthropic Fable 5 for Foreign Users – 72 Hours After Launch

📖 6 min read

Three days. That’s how long Anthropic’s most powerful AI model stayed live before the US government pulled the plug on international access – and took everyone else down with it.

On June 9, 2026, Anthropic launched Claude Fable 5, its first publicly available “Mythos-class” model and the most capable AI the company has ever shipped. By June 12, the Trump Administration had issued an emergency export control directive ordering Anthropic to cut off every foreign national from accessing Fable 5 and its research-tier sibling, Mythos 5 – including Anthropic’s own foreign national employees. Anthropic, unable to surgically enforce the restriction user by user in the time given, did what it called the only viable thing: it disabled both models for all customers worldwide.

It is the first time the US government has weaponized export control law to shut down access to a commercial AI model. The fallout is still unfolding.

What Is Claude Fable 5?

Fable 5 is Anthropic’s first model in a new “Mythos-class” capability tier, positioned above the existing Opus family. It launched June 9 and was immediately available via the API, Claude apps, Amazon Bedrock, and included free for Claude Pro, Max, Team, and Enterprise subscribers through June 22.

The numbers that made it a big deal:

Benchmark	Claude Fable 5	Claude Opus 4.8	Mythos Preview
SWE-Bench Pro (agentic coding)	80.3%	69.2%	~74%
ExploitBench (security reasoning)	78%	40%	69%

Source: Independent third-party testing compiled by Vellum AI and The Decoder, June 2026. Anthropic has not published official benchmark numbers. Treat as directional signals.

The headline capability Anthropic pushed hardest: Fable 5 can run autonomous coding tasks for extended periods – potentially days – with minimal human intervention. It’s priced at $10 per million input tokens / $50 per million output tokens, exactly double Opus 4.8. Sensitive queries are rerouted to the safer Opus 4.8 rather than refused outright, a design Anthropic calls “defense in depth.” More than 95% of Fable sessions, according to Anthropic’s early data, involve no such fallback at all.

This is also the first time Anthropic publicly released a model in its “Mythos” class. Claude Mythos 5 – the same underlying model with some safeguards lifted – was made available only through the restricted Project Glasswing program for infrastructure providers and vetted cybersecurity researchers.

What the Government Ordered – and Why

At 5:21 PM ET on Friday, June 12, Anthropic received a directive from the US government citing national security authorities. The order: suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States. Anthropic said the letter did not explain the specific national security concern in detail.

Anthropic’s understanding, based on verbal briefings it received, is that the government believes someone has found a method to jailbreak Fable 5 – specifically, a technique involving asking the model to read a codebase and identify software vulnerabilities. The government demonstrated the technique to a third party and shared that information with officials.

Anthropic pushed back hard in its public statement, arguing:

• The demonstrated jailbreak is narrow and non-universal – it can elicit some information in specific circumstances, not broadly bypass safeguards
• Other publicly available models, without any jailbreaking, can find the same minor vulnerabilities
• No tester has found a universal jailbreak – one that broadly unlocks the model’s full capabilities
• Anthropic explicitly stated at launch that perfect jailbreak resistance is impossible for any model provider
• The vulnerabilities disclosed are “relatively simple” and provide “no Mythos-specific uplift”

“We have not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result,” Anthropic wrote. “The potential jailbreaks that have been disclosed to us are either entirely benign responses or are minor findings that provide no Mythos-specific uplift.”

Despite the disagreement, Anthropic complied. It could not technically restrict access only to foreign nationals fast enough to satisfy the directive, so it shut both models down for everyone. Access to all other Anthropic models – Opus 4.8, Sonnet, Haiku – was unaffected.

Why This Matters Beyond Anthropic

This is new territory. Export control laws have long governed physical technology – semiconductors, weapons systems, chemical precursors. Applying them to software model weights or API access is a significant legal and policy escalation. The implications:

For AI companies: If the government can flip a switch on one model, it can flip it on any model. OpenAI, Google, Meta, and every other frontier lab now has a direct data point that their most capable products could be restricted – or pulled – by executive action, without detailed justification, with hours of notice.

For international users and developers: Developers in Europe, Asia, and elsewhere who had integrated Fable 5 into products – including, presumably, the paid enterprise customers Anthropic was targeting – lost access overnight. No warning. No workaround. The 2x-priced flagship model simply stopped working.

For the AI safety debate: The situation cuts both ways. On one hand, it demonstrates that the government takes AI-enabled cyber capabilities seriously enough to act on. On the other, Anthropic’s detailed rebuttal – arguing the jailbreak is trivially narrow and matched by existing public models – raises real questions about whether the intervention was proportional or technically informed.

For compute and export controls: The US has spent two years restricting GPU and chip exports to slow AI development in adversarial nations. Cutting off API access is a much blunter instrument – it hits allies, researchers, and paying customers equally with adversaries. That’s a policy tradeoff worth watching.

What Comes Next

Anthropic has not said when – or whether – Fable 5 and Mythos 5 will return. The company said it is “working to understand the government’s concerns” and that it received only verbal evidence of the potential jailbreak. There is no public timeline for restoration.

The model was less than 72 hours into a 14-day free trial window for subscribers. Users who already paid for the more expensive tiers will want answers about whether credits will be extended or refunded.

Separately, the launch blog post for Fable 5 noted that Mythos 5 access would “expand gradually in coordination with the US government” – suggesting Anthropic already anticipated some degree of government involvement in how the model rollout would proceed. What no one apparently anticipated was a full shutdown with hours’ notice.

BetOnAI Verdict

This is the most significant AI policy moment since the EU AI Act took effect – and in some ways bigger, because it happened in real time, to a live product, with a single administrative directive.

On the technical question: Anthropic’s rebuttal is credible. A narrow, non-universal jailbreak that other public models can replicate is not obviously a reason to shut down a model globally. The government may have information Anthropic doesn’t, but the explanation given doesn’t hold up to scrutiny on its own.

On the policy question: this changes the calculus for anyone building on frontier AI APIs. “Model access” is now a political variable, not just a technical or commercial one. If you’re building a product on Fable 5 – or whatever comes after it – your risk model needs to include “government shutdown with 24 hours’ notice.” That’s not theoretical anymore.

For individual users: Claude Opus 4.8, Sonnet, and Haiku all remain accessible. Fable 5 was impressive for 72 hours. Whether it comes back in its current form, or returns with additional restrictions baked in, is now a policy negotiation as much as an engineering one.

The AI race just became a geopolitical one in a much more direct way. Watch this space.

---

Sources:

• Anthropic official statement on US government directive (June 12, 2026)
• Anthropic: Claude Fable 5 and Mythos 5 launch post (June 9, 2026)
• New York Times: U.S. Bars Foreigners From Using Anthropic’s Most Advanced A.I. Models
• Bloomberg: Anthropic Says US Orders Halt to Foreign Access for Fable 5, Mythos 5
• CNBC: Anthropic disables access to Fable 5 and Mythos 5 to comply with government directive
• Vellum AI: Claude Fable 5 and Mythos 5 Benchmarks Explained
• Codersera: Claude Fable 5 – Benchmarks, Pricing & What’s New