Anthropic disabled Claude Fable 5 and its sibling model Mythos 5 for all customers on June 12, 2026, after the US government issued an export control directive citing national security authorities. The order itself does not ban everyone: on its face it bars access by foreign nationals, including Anthropic's own foreign-national staff. But because Anthropic's shared cloud cannot cleanly separate foreign users from domestic ones, the company pulled both models entirely. Anthropic says it is complying while disagreeing, and is working to restore access. No restoration date has been given.
Anthropic published a statement on June 12, 2026 confirming that "the US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5." The company says it received the directive "today at 5:21pm (ET)" and disabled both models the same day.
The framing matters, because the directive is narrower than the outcome. As written, the order prohibits access by "any foreign national, whether inside or outside the United States, including foreign national Anthropic employees." It is not, on its face, a blanket shutdown for every customer. The reason the entire world lost access to Claude Fable 5 is that Anthropic decided full compliance required pulling the models outright.
Anthropic runs Fable 5 and Mythos 5 on shared cloud infrastructure that cannot cleanly partition foreign-national users from domestic ones. With no reliable way to honor a foreign-national-only restriction at the serving layer, the company concluded the only way to comply was to disable both models for all customers. So the total outage is a downstream consequence of Anthropic's compliance decision, made under protest, rather than the literal text of the order.
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5.
Anthropic
On the government side, Bloomberg, CNBC, and NBC News reported that the directive came through the US Department of Commerce's Bureau of Industry and Security, in a letter signed by Commerce Secretary Howard Lutnick to Anthropic CEO Dario Amodei. Anthropic's own statement names only "the US government," not Lutnick or BIS, so those specifics rest on the reporting rather than the primary document.
Claude Fable 5 and Mythos 5 launched on June 9, 2026, three days before the directive. We covered the rollout in our Claude Fable 5 release breakdown. The two are the same underlying model. What differs is the safety layer: Fable 5 ships with safety classifiers in place, while Mythos 5 has those classifiers lifted.
Mythos 5 was not a general consumer release. It was distributed through Project Glasswing, a collaboration with the US government aimed at cyberdefenders and critical-infrastructure providers. The idea was to give vetted defensive teams a less-restricted model for security work, where the standard classifier guardrails can get in the way of legitimate analysis. An earlier Mythos preview had already gone out to roughly 150 organizations under Project Glasswing around April 2026.
That history sets up the central irony of this story. A model line built partly around a US-government cyberdefense partnership was suspended by a US-government directive three days after its public launch. The same capability framing that justified the government collaboration, advanced help with software and security problems, is close to the capability the government now cites as the risk.
According to Anthropic, the government "believes it has become aware of a method of bypassing, or 'jailbreaking' Fable 5." Anthropic says it reviewed the demonstration and reached a very different conclusion: that it revealed only "a small number of previously known, minor vulnerabilities," and that other publicly available models can surface the same issues without any bypass at all.
The specific technique has not been confirmed by Anthropic. Axios, citing an anonymous administration official, reported that Commerce acted after another, unnamed company claimed it had jailbroken Mythos, and that the technique reportedly "involves asking the model to read a specific codebase and fix software flaws." That description maps onto the cyberdefense use case Mythos was built for, which is part of why Anthropic frames the finding as a known capability rather than a novel exploit.
Anthropic's rebuttal rests on the scale of its testing. The company says it red-teamed Fable's safeguards for thousands of hours in total, working with the US government, the UK AI Safety Institute, multiple third-party organizations, and internal teams. It states plainly that "no testers have yet been able to find a universal jailbreak."
The deeper argument is about what standard to hold a model to. Anthropic says perfect jailbreak resistance "is not currently possible for any model provider," and that its layered defenses make the residual risk "comparable to existing models." In other words, the company is not claiming Fable 5 is unbreakable. It is arguing that a narrow, non-universal bypass of the kind found here is the normal state of every frontier model, not a reason to recall a commercial product.
For practical purposes, every customer lost access to Fable 5 and Mythos 5 on June 12, 2026, regardless of nationality. Even though the directive is written around foreign nationals, the shared-infrastructure compliance choice means there is no domestic-only carve-out you can currently reach. If your application or workflow called either model, those calls stopped working that day.
The damage is contained to those two models. Anthropic states that "access to all other Anthropic models will not be affected." That keeps the rest of the lineup online, including Opus 4.8, Sonnet, and Haiku. Builders who depended on Fable 5 will need to fall back to one of these or pause the affected feature until access is restored.
| Model | Status | Notes |
|---|---|---|
| Claude Fable 5 | Disabled | Suspended for all customers; the directive names foreign nationals, the full outage is Anthropic's compliance choice |
| Claude Mythos 5 | Disabled | Same underlying model with classifiers lifted; distributed via Project Glasswing |
| Claude Opus 4.8 | Available | Unaffected by the directive |
| Claude Sonnet | Available | Unaffected by the directive |
| Claude Haiku | Available | Unaffected by the directive |
Anthropic's posture is compliance under protest. It is following the order while openly contesting the reasoning behind it. The company says it is complying with the government's legal directive, and in the same statement says it disagrees that the finding of a narrow potential jailbreak should be cause for recalling a commercial model.
Most of the statement is spent making the case that the bar being applied here is one no current model could clear. The thousands of hours of red-teaming, the involvement of the US government and the UK AI Safety Institute, and the absence of any universal jailbreak are all marshalled to support a single point: a non-universal bypass is not a defect unique to Fable 5, it is the present reality of frontier AI.
From there Anthropic escalates to an industry-wide warning. If the standard used to suspend Fable 5 were generalized, the company argues, the consequences would reach far beyond its own products.
If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.
Anthropic
That is the core of the precedent argument. The claim is not that Fable 5 should be exempt from scrutiny, but that recalling a deployed commercial model over a minor, already-known bypass sets a threshold every frontier lab would fail, every time. Whether regulators accept that framing is the open question, and it is the part of this episode most likely to shape how future model releases are handled.
Anthropic says it is working to restore access to Fable 5 and Mythos 5. It has not committed to a date, and nothing in the statement suggests a quick reversal. Treat the timeline as unknown rather than imminent, and do not read "working to restore" as a promise that the models return on a schedule.
Practically, restoration likely depends on resolving the dispute itself: either the government revises or lifts the directive, or Anthropic builds a way to enforce a foreign-national restriction at the serving layer so it can comply without a total shutdown. The statement does not say which path is in play, so any specific mechanism here would be speculation.
Two threads are worth tracking from here. First, the Project Glasswing irony is not just a talking point. A model line developed in part for a US-government cyberdefense program, detailed in the June 9 launch announcement, was pulled by a US-government export control directive within three days. That tension between collaboration and restriction is unresolved.
Second, this is the first high-profile instance of national security authorities being used to suspend a deployed frontier model in this way. How it ends, whether through a quiet restoration, a formal challenge, or a new compliance carve-out, will set expectations for the next launch that draws government scrutiny. For now, the facts are narrow: Fable 5 and Mythos 5 are down for everyone, the rest of the Claude lineup is up, and there is no return date.