For about two and a half weeks in June 2026, two of the most capable AI models on the planet simply went dark for everyone outside the United States. Not because of an outage, a price change or a policy tweak — but because the US government ordered it. Days after Anthropic launched Claude Fable 5 and Mythos 5, the Commerce Department issued an export-control directive suspending access for any foreign national, anywhere in the world. If you were a UK business relying on those models, they vanished overnight, and there was nothing you could do about it.
It’s one of the strangest episodes in the short history of commercial AI, and it ended almost as abruptly as it began — the controls were lifted on 1 July. But the drama matters well beyond the two companies at its centre, because it exposed a risk most UK teams have never priced in: the AI model you build on can be switched off by a foreign government, and your nationality can put you on the wrong side of the switch. Here’s what actually happened, told straight, and what it means for anyone in the UK depending on US AI tools.
A note on sourcing: this is a contested story with several disputed claims, and we’ve attributed each to whoever made it. It’s a snapshot as of 16 July 2026, drawn from Anthropic’s own statements plus reporting by Fortune, Forbes, CNBC, Al Jazeera, Semafor, Tom’s Hardware and The Hacker News. Where the government and Anthropic disagree, we say so rather than pick a side.
What happened: a two-and-a-half-week blackout
The bare timeline is remarkable enough on its own.
| Date (2026) | Event |
|---|---|
| 12 June | Anthropic publicly launches Claude Fable 5 and Mythos 5, its most capable models to date |
| Mid-June (~13–15 June) | US Commerce Department issues an export-control directive; Anthropic disables both models for all foreign nationals |
| Mid–late June | Public dispute plays out between Anthropic and the White House over the reason and the process |
| 30 June | Commerce notifies Anthropic it has removed the export controls |
| 1 July | Anthropic begins restoring global access, with additional cybersecurity safeguards |
In Anthropic’s own words, posted publicly at the time, the government “citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.” Read that again: it applied to foreign nationals inside the US too, and even to Anthropic’s own overseas staff. The practical effect was a global shutdown of both models for everyone who wasn’t a US person — a category that includes essentially every UK user.
Why did the US do it? Two competing explanations
This is where the story gets murky, because there are two overlapping reasons in circulation, and the parties don’t agree on which was decisive.
Explanation one: the Fable 5 jailbreak. The trigger, according to White House AI adviser David Sacks, was a “jailbreak” — a way of bypassing the model’s safety guardrails. Sacks said a partner trusted by both Anthropic and the government surfaced a bypass of Fable’s guardrails, that the administration asked Anthropic’s chief executive Dario Amodei to fix it or de-deploy the model, and that Amodei declined. The fear, as reported, was that an unguarded frontier model could be turned into an unrestricted cyber-weapon.
Explanation two: Chinese access to Mythos. Separately, Semafor reported that the White House move was linked to concerns that a China-linked group had accessed Mythos, raising the spectre of the model being reverse-engineered or distilled by a foreign adversary — the classic national-security rationale behind US export controls on advanced technology.
Crucially, Anthropic disputes that the two were connected. A company spokesperson said the White House “didn’t raise Chinese access to Mythos in its conversations around the Fable jailbreak” — implying the China angle was a post-hoc justification rather than the actual cause. Whether the ban was really about a security flaw, a geopolitical worry, or a bit of both, remains genuinely unclear.
Anthropic’s rebuttal
Anthropic did not go quietly. Its public position had three parts, and they’re worth understanding because they frame a real debate about how AI should be governed.
- The jailbreak was narrow, not catastrophic. Anthropic characterised the bypass as non-universal — essentially getting the model to read a codebase and identify software flaws — and argued the same result could be produced on other public models, including OpenAI’s GPT-5.5. In other words, this wasn’t a unique superweapon.
- The remedy was disproportionate. The company said it disagreed that a narrow jailbreak should justify recalling a commercial model already used by hundreds of millions of people. If that became the standard, it warned, it “would essentially halt all new model deployments for all frontier model providers” — nobody could ever ship, because every frontier model has some jailbreak.
- The process was opaque. Anthropic criticised the lack of a specific, evidenced reason and the speed with which a national-security authority was used against a commercial product.
You don’t have to take Anthropic’s side to see the point: if governments can pull any model that can be jailbroken — and all of them can be, to some degree — the bar for shipping frontier AI becomes almost impossible to clear.
How it ended: conditions, not capitulation
On 30 June, the Commerce Department notified Anthropic that it had removed the export controls, and access began returning from 1 July. This wasn’t a simple climbdown by either party — it came with commitments. Commerce Secretary Howard Lutnick said Anthropic would no longer require an export licence because it had agreed to:
- proactively detect and address security risks associated with the models;
- work with the government on standards for its upcoming models; and
- inform the government of malicious activity.
Anthropic restored the models with additional cybersecurity classifiers it described as among its strongest safeguards. So both sides got something: the government secured an ongoing oversight relationship and new commitments; Anthropic got its flagship models back on the market worldwide. Whether that’s a healthy precedent or a worrying one — a government and a lab quietly negotiating the terms on which a model may exist — is a question the industry will be chewing over for a long time.
This episode also lands right as the wider AI landscape is shifting fast: it came just weeks before OpenAI’s own GPT-5.6 launch, which itself waited on a US government review before broad release. Government sign-off is quietly becoming part of the frontier-model release cycle.
The bit that matters for UK teams
Here’s the uncomfortable takeaway, and it has nothing to do with the rights and wrongs of the jailbreak. For two and a half weeks, UK users were collateral. The directive didn’t target the UK; it targeted “foreign nationals,” and from Washington’s perspective, that’s us. A British company that had built a product, a workflow or a customer-facing feature on Fable 5 or Mythos 5 lost access with no notice, no appeal and no UK involvement in the decision.
That is a category of risk most teams never model. We spend a lot of time on this site worrying — rightly — about where your data goes under UK GDPR and data residency. This is the mirror-image problem: not where your data sits, but whether the model itself will still be available to you tomorrow, given that the company providing it answers to a foreign government. Continuity, not just compliance.
None of this makes Claude a bad choice — it remains our top pick for coding and hard reasoning, and Anthropic arguably came out of this looking principled. But it’s a sharp reminder that “US AI model” carries a geopolitical dependency that a UK buyer should plan around.
What to actually do about it
You can’t control US export policy. You can control how exposed your organisation is to it. Practical steps:
- Don’t put a single US model on a critical path with no fallback. If a core process would break the moment one provider goes dark, that’s a single point of failure — regardless of how good the model is.
- Build for model-portability. The maturing Model Context Protocol and the general convergence of model APIs make it more realistic than ever to swap providers. Design workflows that aren’t welded to one vendor, so you can fail over from, say, Claude to ChatGPT or Gemini if you must. Our ChatGPT vs Claude vs Gemini comparison is a good starting point for picking a viable second source.
- Keep a self-hosted or open-weight option in the mix for the truly sensitive stuff. A model running on infrastructure you control can’t be remotely switched off by anyone’s government.
- Read the availability terms, not just the privacy terms. Ask a prospective vendor what happens to your access under a government order, an outage or a country-level restriction. Most contracts are silent on it — which is itself an answer.
- Log which models touch which processes. When Fable 5 went dark, the teams that recovered fastest were the ones that knew exactly where they’d deployed it. You can’t fail over a dependency you haven’t mapped.
Frequently asked questions
Why did the US ban Claude Fable 5 and Mythos 5? Officially, on national-security grounds under export-control authorities. Reporting points to two overlapping reasons: a “jailbreak” of Fable 5’s safety guardrails (per White House adviser David Sacks), and concerns that a China-linked group had accessed Mythos (per Semafor). Anthropic disputes that the two were connected and argues the jailbreak was narrow and reproducible on other models.
How long were the models unavailable? Roughly two and a half weeks. They launched on 12 June 2026, were suspended for foreign nationals within days, and access was restored from 1 July after Commerce lifted the export controls.
Did this affect UK users? Yes. The directive suspended access for all “foreign nationals” worldwide, which includes UK individuals and businesses. Anyone in the UK using Fable 5 or Mythos 5 lost access during the suspension.
Are Fable 5 and Mythos 5 available now? Yes. Anthropic began restoring global access on 1 July 2026, with additional cybersecurity safeguards, after agreeing to detect and report security risks and to work with the US government on standards for future models.
Does this mean I shouldn’t use Claude? No. Claude remains an excellent, and in some areas leading, family of models. The lesson isn’t “avoid Claude” — it’s “don’t build a critical process on any single US model with no fallback,” because availability can be affected by decisions taken outside your control.
Final thoughts
The Fable 5 and Mythos 5 saga will be remembered less for the jailbreak at its centre than for the precedent around it: a frontier AI model, live to hundreds of millions of people, switched off for the rest of the world by government order — then switched back on under negotiated terms. For UK organisations, the moral isn’t to distrust any one company. It’s to recognise that frontier AI now sits at the intersection of technology and geopolitics, and to build with the resilience that implies: a second source, a portability plan, and eyes open about who ultimately controls the off switch.
Want help building an AI stack that won’t fall over when the geopolitics do? Get in touch — we’re always happy to help.
Last updated: 16 July 2026. This account involves contested claims, each attributed to its source; figures and quotes were drawn from Anthropic’s own statements plus reporting by Fortune, Forbes, CNBC, Al Jazeera, Semafor, Tom’s Hardware and The Hacker News. See our editorial standards.