Claude Fable 5 was cut off for everyone on 12 June 2026 on the direct order of the US government, for cybersecurity reasons. After 18 days of negotiations, the restrictions were lifted on 30 June, and the model becomes accessible worldwide again from today, 1 July. Here is the full story of this unprecedented episode.
There is news that is announced with great fanfare, and news that lands like a guillotine. On Friday 12 June 2026, at 5:21pm New York time, it was the second category that hit Anthropic. The US government had just served it with an export control directive, forcing it to cut access to its two most advanced models, Fable 5 and Mythos 5, worldwide. Today, 1 July, Fable 5 officially returns. Here is what happened between those two dates, and why this episode will leave a lasting mark.
Back to the cause: an Amazon report
To understand the trigger, you have to go back to the initial launch. Fable 5 and Mythos 5 were released on 9 June 2026. They share the same underlying model, but Fable 5 was released with solid guardrails for general consumer use, while Mythos 5, with fewer restrictions, was only entrusted to a small number of trusted partners in the Glasswing programme, for defensive cybersecurity.
The 12 June directive came after the government learned of a report from Amazon researchers, who had found a method to bypass (known as "jailbreaking", literally "escaping prison") Fable 5's guardrails by phrasing prompts in a way that made it identify software vulnerabilities. In one case, the model had even produced code demonstrating how to exploit the flaw in question.
Anthropic conducted its own investigation into this report and discovered something crucial: the vulnerabilities identified were all relatively simple and already known. What's more, far less capable models, including Claude Opus 4.8, GPT-5.5 and even the Chinese model Kimi K2.7, could spot exactly the same flaws without any jailbreak at all. The problem was therefore not specific to Fable 5, but Fable 5 paid the price of being the model in the spotlight at the wrong time.
18 days of uncertainty and negotiation
The cut-off was brutal and left no room for manoeuvre: the directive took effect immediately, and Anthropic had no reliable way to verify the nationality of its users in real time. As a result, the company had to suspend access to both models for everyone, including its own employees of foreign nationality, as it was unable to isolate only the users covered by the ban.
The following weeks were marked by growing frustration, born of an almost absurd collision of events. Even before the cut-off, Anthropic had promised its Pro, Max, Team and Enterprise subscribers free access to Fable 5 for 13 days, from 9 to 22 June, before a planned switch to a paid credit system on 23 June. Except the model remained suspended for absolutely everyone from 12 June through to 30 June inclusive, with no restoration whatsoever between those two dates: API calls returned an error continuously throughout that period. As a result, the 22 June deadline arrived while Fable 5 was still offline, meaning the vast majority of subscribers never got a single hour of their free window before it expired on paper. Several developers were publicly demanding compensation for this free trial that had evaporated before it had even begun.
On 26 June, a first political gesture arrived: the government authorised Mythos 5 to be redeployed for a set of US organisations operating critical infrastructure. But Fable 5 remained blocked, with no return date announced. It would take another four days.
The outcome: a cooperation agreement
On 30 June 2026, Anthropic announced that the export restrictions on Fable 5 and Mythos 5 had been lifted. In a letter addressed to Anthropic co-founder Tom Brown, US Commerce Secretary Howard Lutnick explained that the decision reflects "Anthropic's close coordination and cooperation with government officials" in addressing the risks associated with its models.
In exchange, Anthropic commits to continuing to collaborate with the government "on protocols, standards and releases for Mythos, Fable, and future models", and to informing the state of any malicious activity detected. In other words, Anthropic accepts greater government oversight of its future releases of frontier models, in exchange for the lifting of the ban.
What actually changes from today
From today, 1 July 2026, Fable 5 becomes accessible worldwide again, without nationality restrictions, on the Claude platform, Claude.ai, Claude Code and Claude Cowork. This is an entirely new window, reset from scratch: it has nothing to do with the June one, wiped out by the cut-off. For Pro, Max, Team and some Enterprise plans, Fable 5 is included free up to 50% of weekly usage limits, and that lasts until 7 July, after which you'll need to move to a paid credit system. Access via AWS, Google Cloud and Microsoft Foundry, however, does not follow exactly the same schedule: Anthropic promises to restore it "as quickly as possible", without committing to a specific date, which suggests a delay of a few days for businesses going through these cloud platforms rather than Claude directly.
One nuance to keep in mind if you don't see anything change on your end today: until the eve of this announcement, several independent industry observers were suggesting the possibility of a gradual rollout, account by account, rather than an instant switch for all users at once. The schedule announced by Anthropic is clear and dated, but a slight delay of a few hours on your specific account would not be unusual.
Anthropic is also taking the opportunity to propose, with Amazon, Microsoft, Google and other Glasswing programme partners, a common framework for rating the severity of jailbreak attempts across the entire industry. An attempt to turn an isolated crisis into a shared standard, to prevent the same episode from happening elsewhere.
Why this episode matters, beyond Anthropic
This affair marks a quiet but real turning point. It is the first time a government has forced the immediate, worldwide shutdown of an already-deployed commercial AI model for reasons of national security. Until now, AI regulation was mostly upstream (chip export controls, limits on who can buy what). Here, the state acted directly on a product already in service, with immediate effect.
The precedent carries heavy consequences for the entire industry. A few days after this affair, OpenAI launched GPT-5.6 under a deliberately restricted access regime, in consultation with the same government, proof that the message has been received across the sector. The lesson many technical leadership teams are drawing: a frontier model accessible today can be cut off tomorrow, without much notice, on a simple national security decision. Predictability of access is becoming as important a selection criterion as raw performance.
This cut-off came a few days after Anthropic itself denounced, in its letter to the Senate on the Alibaba affair, the fact that Chinese players had siphoned off Claude's capabilities undetected for six weeks. The Fable 5 episode shows that vigilance is not only directed outward: the US government judged that Anthropic itself needed closer monitoring. No one is beyond the reach of Washington's gaze, not even those who call for it to be applied to others.
Eighteen days is short on the scale of a lifetime, but long on the scale of 2026's AI news cycle, where everything plays out in days and sometimes hours. Fable 5 returns today more watched, more regulated, but also more legitimate: it has survived the first real test of trust between a government and an already-deployed frontier model. The open question is whether this precedent will remain the exception, or whether it becomes the new norm for all future model releases.