US Government Forces Anthropic to Pull Claude Fable 5 and Mythos 5 Worldwide Over Export Control Directive
A Commerce Department directive citing national security concerns forced Anthropic to suspend all access to Claude Fable 5 and Mythos 5 globally.
At 5:21 PM ET on June 12, Anthropic received a directive from the US Commerce Department ordering it to immediately suspend all access to Claude Fable 5 and Claude Mythos 5 by any foreign national. Because Anthropic cannot reliably determine which users are foreign nationals in real time, the only practical way to comply was to take both models offline for everyone, worldwide. That is exactly what the company did.
All other Claude models remain available. But if you were using Fable 5 or Mythos 5, you just lost access to them with no warning, and no clear timeline for getting them back.
What happened, exactly?
Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei, citing national security concerns and invoking export control authorities. The directive was prepared with input from the Commerce Department’s Bureau of Industry and Security, the same body that oversees chip export restrictions. This appears to be the first time those tools have been turned on a commercially deployed language model.
According to Axios, the administration was already uncomfortable with Anthropic’s plans to release the new models. When Anthropic went ahead with the launch on June 9 despite pressure to pause, the export control letter followed three days later.
The stated trigger was a claimed jailbreak of Mythos 5, reported to the administration by another company. The concern is that Mythos 5, which operates with some of Fable 5’s safety filters removed for a vetted set of cybersecurity organisations, could be used to cause serious harm if someone bypassed those remaining controls.
Anthropic’s position
Anthropic is complying with the directive. It is not, however, agreeing with the reasoning behind it.
In its public statement, the company outlined several objections. First, the government provided only verbal evidence of the jailbreak, and the technique described amounted to asking the model to read a codebase and identify software flaws. When Anthropic reviewed the specific demonstration, it found the output contained only a small number of previously known, minor vulnerabilities. The same output, the company says, can be produced by OpenAI’s GPT-5.5 without any bypass at all.
Second, Anthropic argues that if a partial, non-universal jailbreak is sufficient grounds to recall a commercial model deployed to hundreds of millions of people, that standard would effectively halt all new model deployments across the entire industry. No frontier model in wide deployment is immune to narrow jailbreaks.
Third, the company says the directive came without a transparent statutory process, without specific technical details, and without the kind of fairness it would expect from a regulatory action of this magnitude.
Anthropic believes this is a misunderstanding and says it expects to share more technical details within 24 hours of the suspension.
What is Fable 5, and why does it matter?
Anthropic launched Fable 5 on June 9 as its most capable publicly available model. It is built on the same underlying architecture as Mythos 5, with classifiers added to block responses in high-risk areas including cybersecurity, biology, and chemistry. Those filters redirect sensitive queries to Claude Opus 4.8 instead, which Anthropic says happens in fewer than 5% of sessions.
Mythos 5 sits alongside it as a more capable variant, deployed through Project Glasswing in collaboration with the US government, and accessible only to a separately vetted group of cyber defenders and infrastructure providers. Ironically, the model was built partly in partnership with the government that has now ordered it offline.
Before launch, Anthropic ran thousands of hours of red-teaming with the US government, the UK AI Safety Institute, and multiple private third-party organisations. None of those tests produced a universal jailbreak.
What does this mean for you?
If you are an individual user or a business that started building on Fable 5 after the June 9 launch, you are now without access and no one has told you when it is coming back.
That uncertainty is the real practical problem here. Businesses that integrated Fable 5 into workflows or products over the past three days are now blocked globally, not because of anything they did, but because a regulatory letter arrived on a Friday afternoon. Microsoft had already restricted employee access to the model over data retention concerns, so some enterprise customers were already navigating a complicated picture before this suspension added another layer.
The broader signal is harder to ignore. Export control law was designed for hardware, components, and military technology. Applying it to a language model already running in commercial production, used by hundreds of millions of people, is genuinely new territory. If the Commerce Department can pull a model this widely deployed on the basis of a narrow jailbreak claim that the developer disputes, every AI company with a frontier model needs to think about what that means for them.
Anthropic has said it is working to restore access as quickly as possible. For now, Claude Opus 4.8 and other models in the Claude family are still running normally. If Fable 5 was central to something you are building, it is worth checking Anthropic’s news page directly for updates over the next 24 to 48 hours.
Updates to this story
7 July 2026: Claude Fable 5 subscription inclusion ends, moving to usage credits at $10/$50 per million tokens
The export controls covered in the original article were lifted on June 30, and Claude Fable 5 returned globally on July 1 across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork. A short grace window included Fable 5 in Pro, Max, Team, and select Enterprise weekly limits at up to 50% of the normal allowance. That window closes today.
From July 8, Fable 5 is no longer included in any subscription tier. All access moves to prepaid usage credits billed at $10 per million input tokens and $50 per million output tokens. Standard Enterprise seats have no included allowance at all. Critically, there is no automatic fallback to another model if credits are not pre-enabled, meaning users without credits simply lose access.
Anthropic describes this as a temporary capacity measure rather than a permanent pricing change, citing unpredictable demand. A new safety classifier is also active; blocked requests are rerouted to Opus 4.8 at no Fable-rate charge, though the classifier produces more false positives during routine coding tasks.
If you have Fable 5 in any workflow today, enable usage credits now in your account or team Settings to avoid an access gap. For bulk or production coding, routing to Sonnet 5 at roughly one-fifth the per-token cost remains the practical alternative.
Sources: Anthropic, Redeploying Claude Fable 5 and Claude Fable 5 product page.
1 July 2026: Claude Fable 5 restored globally after US export controls lifted — plus a new industry jailbreak severity framework
The suspension described in this article is over. The US Commerce Department lifted its export controls on Anthropic’s advanced models, and Anthropic began restoring access to Claude Fable 5 on July 1 across Claude.ai, Claude Code, Claude Cowork, and the Claude Platform, with AWS, Google Cloud, and Microsoft Foundry access to follow. Mythos 5 remains restricted but is available to select US organizations following government approval granted on June 26.
In exchange for the controls being lifted, Anthropic agreed to proactively detect security risks, report malicious activity to the government, and provide designated government partners with pre-release access to future frontier models for independent evaluation before broad deployment.
The jailbreak that contributed to the original suspension has also been addressed. A new automated safety classifier now blocks the reported technique in over 99% of cases. Blocked requests are redirected to Opus 4.8. The classifier does increase false positives on some coding and debugging tasks, and Anthropic says it will continue refining it.
Separately, Anthropic is working with Amazon, Google, and Microsoft on a proposed industry-wide framework for scoring jailbreak severity across four dimensions: capability gain, breadth, weaponizability, and discoverability. Through July 7, usage of Fable 5 is capped at 50% of weekly limits for most paid plans.
Full details are in Anthropic’s official post.
30 June 2026: US lifts export controls on Claude Fable 5 — global access restores July 1
The Commerce Department lifted its June 12 export control order on June 30, ending the 18-day suspension described in this article. Fable 5 returns to Claude.ai, the API, Claude Code, and Claude Cowork on July 1. AWS, Google Cloud, and Microsoft Foundry access will be re-enabled as soon as possible.
The original directive was triggered by a report from Amazon researchers describing a technique that caused Fable 5 to identify software vulnerabilities. Anthropic responded by training a new safety classifier that blocks the specific behavior in over 99% of cases. Requests caught by the classifier are rerouted to Opus 4.8 at no Fable 5 pricing. (Anthropic)
To have the export license requirement dropped, Anthropic agreed to proactively detect and address security risks, coordinate with the government on future release protocols, and report malicious activity found in its models.
Mythos 5 remains restricted to approved US organizations following a separate government clearance granted June 26.
Pro, Max, Team, and select enterprise subscribers receive up to 50% of weekly usage limits in Fable 5 credits through July 7. Standard Enterprise seats have no included allowance and require usage credits to be enabled. Premium Enterprise seats include Fable 5 within existing seat usage through July 7.
The broader question of whether the US government will require approval for future frontier model releases remains open. (CNBC)
24 June 2026: Anthropic accuses Alibaba of running the largest-ever distillation attack on Claude
The Fable 5 and Mythos 5 suspension now has a clearer origin. In a letter dated June 10, 2026, addressed to Senate Banking Committee leaders Tim Scott and Elizabeth Warren, Anthropic accused operators linked to Alibaba’s Qwen AI lab of running the largest known distillation attack in the company’s history. The campaign used nearly 25,000 fraudulent accounts to generate 28.8 million Claude exchanges between April 22 and June 5, 2026, one week before the Commerce Department directive pulled both models offline.
Model distillation does not involve stealing weights or source code directly. It uses outputs from a target model to train a rival system, allowing competitors to replicate capabilities at a fraction of the development cost. Anthropic says the operation focused specifically on software engineering and agentic reasoning, the capabilities central to Mythos Preview. Reuters reviewed the letter; Alibaba has not yet responded publicly, and its stock fell nearly 3 percent on Wednesday.
This disclosure reframes the original story. What the earlier post described as a national security export control action now appears to be directly connected to a documented foreign intelligence-gathering campaign. Anthropic is urging Congress to expand threat intelligence sharing between the government and private AI laboratories. The White House flagged distillation as a national security concern in April 2026, when OSTP Director Michael Kratsios issued related guidance to US AI labs.
23 June 2026: Claude Fable 5 moves behind a usage-credit paywall for all subscribers as complimentary window ends
As of June 23, Fable 5 is no longer included in Claude Pro, Max, Team, or Enterprise subscriptions. Any continued use draws from prepaid usage credits billed at API rates: $10 per million input tokens and $50 per million output tokens, double the cost of Opus 4.8. For context, a developer running roughly two hours of agentic Fable 5 work per day faces approximately $150 per month in new costs; a 100-developer team that does not adjust usage faces around $15,000 per month in overages.
This adds a significant financial layer to the original story. Subscribers were offered a 13-day complimentary window starting June 9, but the government-mandated suspension that took effect June 12 wiped out the bulk of that period, leaving most users with only four to five usable days before the credit cutover arrived regardless.
Anthropic has described the credit-only arrangement as temporary, stating it intends to restore Fable 5 as a standard plan feature once capacity allows, though no timeline has been given. As of this update, the government suspension of both Fable 5 and Mythos 5 remains unresolved, meaning the paywall is now live while the underlying access ban is still in effect. All other Claude models remain on existing subscription terms.
17 June 2026: Anthropic executives signal US Fable 5/Mythos 5 export controls likely to be resolved within days
Fable 5 and Mythos 5 remain suspended as of this update. However, remarks from Anthropic’s head of international Chris Ciauri at a Seoul press briefing on June 17 offer the first public timeline signal from the company since the June 12 suspension.
According to DigitalToday, Ciauri said he did not expect the controls to remain in place and indicated a resolution was likely within days. He also stressed that the jailbreak vulnerability at issue was narrow in scope and consistent with what had appeared in models released over the past six months, expressing confidence that Fable 5 and Mythos 5 would become usable again in the near term.
That framing should be treated with caution. UPI and the Korea Times both reported that Ciauri declined to comment on Washington’s restrictions at the same briefing. The accounts conflict, and no official resolution timeline has appeared on the Anthropic newsroom or its existing directive statement.
The Seoul briefing coincided with Anthropic’s formal opening of its Seoul office and a set of new Korean enterprise partnerships. All other Claude models remain unaffected. International enterprises should continue to treat Fable 5 and Mythos 5 access as suspended until Anthropic publishes an official update.
17 June 2026: Anthropic opens Seoul office, signs AI safety MOU with South Korean government, and announces major Korean enterprise deployments
Anthropic formally opened its Seoul office on June 17, 2026, adding important context to the export control story covered below. The expansion arrived less than a week after Washington suspended foreign national access to Fable 5 and Mythos 5, creating a notable tension: Anthropic was simultaneously deepening its Korean market commitments while those commitments were being constrained by US policy.
The Seoul office, led by country head KiYoung Choi, is Anthropic’s first staffed Northeast Asian presence outside Japan. On June 18, Anthropic updated its official announcement to include a signed MOU with Korea’s Ministry of Science and ICT covering public sector AI safety, Korean-language model evaluation, and cybersecurity information sharing.
New enterprise deployments span NAVER, Samsung SDS, LG CNS, and Nexon, among others. SK Telecom, a $100 million Anthropic investor whose access to Mythos through Project Glasswing was reportedly revoked, was not listed among the new deployment partners.
At the Seoul press briefing, Anthropic’s international managing director Chris Ciauri declined to comment on Project Glasswing and said he expected the export controls to resolve within days. No official commitment was made.
Fable 5 and Mythos 5 remain suspended for foreign nationals. The Seoul office opening does not change that. The situation reported below remains unresolved, though Anthropic’s public posture suggests it expects a near-term fix.