The United States government has ordered Anthropic to suspend worldwide access to its two most capable artificial intelligence models, Fable 5 and Mythos 5, citing national security. It is the first time Washington has used export-control authority to pull a commercially deployed American AI model out of service, and it has done so by reaching not at a foreign adversary but at the company’s entire global user base.
Anthropic said it received the directive on Friday, June 12, 2026, at 5:21 p.m. Eastern time and disabled both models the same evening. Every other Claude model remains available. The two that went dark are the company’s flagship release, Fable 5, and its most powerful system, Mythos 5, the latter of which had only ever been offered to a restricted set of customers.
What the directive requires
The order came from Commerce Secretary Howard Lutnick, who addressed a letter directly to Anthropic CEO Dario Amodei on the evening of June 12. According to reporting on the directive, it requires a license for the export, re-export, or domestic transfer of Fable 5 and Mythos 5, and it extends that restriction to any foreign national, including foreign nationals physically inside the United States and Anthropic’s own non-citizen employees. Anthropic was given approximately 90 minutes to comply before disabling the models. [Updated June 2026]
That last point is what makes the order unusual. Conventional export controls stop a product at the border. This one follows the model inside the country and bars a class of people from touching it regardless of where they sit. Anthropic has staff who are not U.S. citizens, and the directive does not carve them out.
Why the government acted
Officials tied the action to the models’ cybersecurity abilities. Mythos 5 is, by Anthropic’s own description, unusually strong at reading software and finding flaws in it, including vulnerabilities that have gone unnoticed for years. Fable 5 ships with safeguards meant to keep ordinary users from reaching that level of offensive capability.
According to reporting, the Commerce Department moved after Amazon researchers shared a report with White House officials showing they had bypassed portions of Mythos’s safeguards that restrict information about cyberattacks. Amazon CEO Andy Jassy called senior administration officials on the Thursday night before the directive — June 11 — to relay the finding, and Lutnick’s letter followed the next evening. [Updated June 2026] The government’s letter, by Anthropic’s account, did not spell out the specific technical concern in detail. Amodei subsequently argued in calls with senior officials that what Amazon demonstrated was narrow rather than a full jailbreak of the model’s safeguards.
Beyond cyber, at least one report indicated the order also pointed to Fable 5’s capabilities in biotechnology, another domain where frontier models draw national security scrutiny. Fable 5 ships broad biology and chemistry classifiers that route flagged prompts to a fallback model — a design Anthropic built explicitly to limit dual-use biology outputs — but those safety classifiers did not appear to satisfy the government’s threshold.
How Fable 5 and Mythos 5 differ
The two suspended models sit at different tiers. Both launched on June 9, 2026 — Fable 5 generally available on the Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry; Mythos 5 restricted to a vetted group. The suspension came three days later. Fable 5 is the public flagship Anthropic released to general customers, the model it put forward on coding benchmarks like FrontierCode, CursorBench, and ViBench. Mythos 5 is the larger, more capable sibling that was never opened to the general public; access was gated to a vetted group under a program the company described in its own access rules.
Fable 5’s guardrails are part of what keep a general-purpose user from reaching Mythos-grade offensive security behavior. That is precisely why a jailbreak of the public model set off alarms about Mythos-level capability leaking into hands the company had tried to keep it away from.
Anthropic says it is a misunderstanding
Anthropic is complying, but it does not agree. In a public statement, the company said it was “removing access to Fable 5 and Mythos 5 for all users” while making clear it considered the recall unjustified.
The company characterized the demonstrated jailbreak as narrow rather than universal. By its account, the technique amounted to asking the model to read code and point out weaknesses, and the flaws it surfaced were minor and already known, the sort of thing competing systems such as OpenAI’s GPT-5.5 can also identify. Anthropic argued that its layered safety design puts the real-world risk of Fable 5 on par with models already in wide use across the industry.
Its broader objection is about precedent. “We believe this is a misunderstanding and are working to restore access as soon as possible,” the company said, adding an apology to customers for the disruption. Anthropic warned that treating a narrow, non-universal jailbreak as grounds for pulling a model used by hundreds of millions of people would, if applied evenly, “essentially halt all new model deployments for all frontier model providers.”
Why a limited order became a global blackout
On paper the directive targets foreign nationals, not the general public. In practice Anthropic concluded it could not enforce that line. The company has no reliable way to verify in real time which of its users around the world are foreign nationals, and the rule reaches even non-citizens standing on U.S. soil. Rather than attempt a partial block it could not guarantee, Anthropic switched the two models off for everyone.
The result is that a measure aimed at controlling who can use a piece of software ended up removing it from the market entirely, at least for now.
A relationship already under strain
The export order did not arrive in a vacuum. It is the sharpest turn yet in a deteriorating relationship between Anthropic and the Trump administration.
The friction traces back to a dispute over how the U.S. military could use the company’s technology. Anthropic resisted allowing its models to be built into fully autonomous weapons systems without safety guardrails, a position that put it at odds with parts of the defense establishment. A separate analysis of the Pentagon’s AI acquisition process found that the DoD Software Acquisition Pathway has no defined milestones for model re-validation or data provenance — a structural gap that may have made a blunt export order more attractive than a negotiated technical fix. In February 2026, the administration directed federal agencies to stop using Anthropic’s products. In March, the Pentagon designated the company a “supply chain risk.” Anthropic, for its part, has taken the administration to court over the blacklisting.
Seen against that backdrop, the suspension of Fable 5 and Mythos 5 reads less like an isolated security call and more like the latest move in a sustained standoff between a leading AI lab and a government that has grown wary of it.
What it means for users and the industry
For customers, the immediate effect is blunt: two models, including Anthropic’s newest and most capable, are simply gone for the moment. Teams that built workflows on Fable 5 have to fall back to earlier Claude models, which remain online, even as a separate June billing change was already reshaping how they budgeted for the model.
For the wider industry, the more important signal is the mechanism. Washington has now shown it is willing to use export-control law not just to keep advanced chips and models away from rival nations, but to reach a U.S. company’s live product and shut it down over a capability concern. Frontier labs whose models are good at finding software vulnerabilities, which is to say most of them, now have to weigh the possibility that a demonstrated jailbreak could trigger the same response.
It also raises an unresolved question about consistency. If Fable 5’s code-analysis abilities are comparable to those of other deployed models, as Anthropic contends, then the standard applied here is one many competitors could also fall afoul of. Whether the government intends to apply it evenly, or whether this episode is bound up in its specific quarrel with Anthropic, is not yet clear.
What happens next
Anthropic says it is working to restore access and frames the whole episode as a fixable misunderstanding. A resolution will likely depend on private exchanges with the Commerce Department over the nature of the jailbreak and whether the company’s guardrails are judged sufficient. A license, a narrowing of the order, or a technical fix to Fable 5’s safeguards could each bring the models back.
As of June 17, prediction markets on Kalshi placed 58% odds on restoration before July 1 and 74% before July 10 — a relatively optimistic read, though those odds also imply a meaningful probability of an outage lasting a month or more. The export dispute is unfolding in parallel with Anthropic’s existing litigation against the Pentagon over the supply chain risk designation, where a federal district court in California has issued an injunction blocking the blacklisting while a D.C. appeals court denied Anthropic’s bid for a separate stay. The two legal tracks are procedurally distinct but politically intertwined: any negotiated resolution of the export order will occur against the backdrop of a company already in active litigation with the administration.
The suspension also has a cascading infrastructure effect that extends beyond access to the models themselves. AWS Bedrock, which had only recently begun offering Mythos-class capabilities, now has neither model available — compounding the data-residency and compliance questions that Bedrock’s own provider data-sharing requirements had already raised for regulated enterprises.
Until then, two of the most capable AI systems available anywhere are offline, pulled not by their maker’s choice but by an order from Washington, and the precedent that order sets is likely to outlast the outage itself.
