- The Anthropic Claude export control order from Commerce Secretary Howard Lutnick has forced both Fable 5 and Mythos 5 offline for all customers worldwide.
- Anthropic says the Anthropic Claude export control directive is based on a narrow jailbreak that even publicly available models like GPT-5.5 can already replicate.
- The pulled models trace back to Claude Mythos Preview, a security-research AI that helped Mozilla alone fix hundreds of vulnerabilities.
- Existing Fable 5 sessions will end with an error; users are being redirected to Opus 4.8 or their chosen default model in the meantime.
- The Anthropic Claude export control order from Commerce Secretary Howard Lutnick has forced both Fable 5 and Mythos 5 offline for all customers worldwide.
- Anthropic says the Anthropic Claude export control directive is based on a narrow jailbreak that even publicly available models like GPT-5.5 can already replicate.
- The pulled models trace back to Claude Mythos Preview, a security-research AI that helped Mozilla alone fix hundreds of vulnerabilities.
- Existing Fable 5 sessions will end with an error; users are being redirected to Opus 4.8 or their chosen default model in the meantime.
Table of Contents
Anthropic Claude Export Control: The Government Steps In
The Anthropic Claude export control order landed like a grenade on Thursday evening. Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei informing him that Claude Fable 5 and Claude Mythos 5 — models the company had only launched days earlier — would immediately be subject to export controls applying to any location outside the United States and to all foreign nationals inside it. The practical consequence was swift and blunt: Anthropic disabled both models for every customer on the platform, regardless of where they were or what they were doing.
Anthropic’s own statement spelled out the timeline with unusual precision: “We received the directive from the government today at 5:21pm (ET).” That kind of timestamp isn’t just administrative detail — it signals a company that wants its users to understand this wasn’t a gradual wind-down. It was an abrupt, same-day shutdown triggered by external authority, not internal choice. The Anthropic Claude export control mechanism used here bypassed any standard regulatory notice period entirely.
What Fable 5 and Mythos 5 Actually Were
To understand why Washington moved so fast, you need to know where these models came from. Both Fable 5 and Mythos 5 are descendants of Claude Mythos Preview, an AI system Anthropic built explicitly for security research. This wasn’t a general-purpose chatbot with a few safety filters bolted on — Mythos Preview was engineered to find software vulnerabilities, and it was reportedly very good at it.
Access to Mythos Preview was tightly restricted under something Anthropic called Project Glasswing, limited to a small circle of vetted companies and research partners. The results were striking. Mozilla alone said it resolved hundreds of vulnerabilities as a direct result of working with the model. That’s not a trivial number. For context, most major bug-bounty programmes celebrate dozens of meaningful fixes per year. Hundreds attributable to a single AI tool in a short pilot window suggests Mythos Preview was operating at a level that genuinely changed what’s possible in automated security research.
When Anthropic brought Fable 5 to public release this week, it came loaded with guardrails. The Mythos family, by contrast, stayed locked inside Project Glasswing. The distinction mattered — Fable was the sanitised, commercially deployable version; Mythos was still the sharp instrument in a controlled environment. Some of Fable 5’s restrictions drew immediate criticism from developers who felt the safety measures were too aggressive for practical use, and Anthropic had already started tweaking those settings before the Anthropic Claude export control shutdown notice arrived.
The Jailbreak That Spooked the Administration
According to Axios, which broke the story, the trigger was a claim from another company that it had successfully jailbroken Mythos 5. That was enough to alarm the administration. The Trump White House had actually tried to block Anthropic from releasing the model in the first place and failed — so when a bypass appeared to surface publicly, the government moved to the next available lever: the Anthropic Claude export control framework.
Anthropic isn’t buying the severity of the threat. The company reviewed a demonstration of the specific jailbreak technique and described what it found in candid terms. The vulnerabilities exposed were described as relatively simple, previously known, and — critically — discoverable by other models already on the market without any bypass required. Anthropic named OpenAI’s GPT-5.5 as one such model capable of the same thing. That’s a pointed observation. If the administration’s concern is that Fable 5 can be manipulated into revealing minor, known vulnerabilities, and GPT-5.5 can do the same without any jailbreak, then the logic for singling out Anthropic under an Anthropic Claude export control order starts to look thin.
The company was direct about the limits of the government’s justification: “The letter did not provide specific details of its national security concern.” That’s a carefully measured sentence, but the implication is clear. Anthropic believes this directive is based on an incomplete picture.
Anthropic Claude Export Control: What Happens to Users Now
For anyone running integrations or active sessions on Fable 5, the experience is messy. Claude’s official account on X confirmed that new sessions will default to the user’s chosen model or Opus 4.8, while any existing Fable 5 sessions will terminate with an error. API integrations pointing at Fable 5 will also return errors. Anthropic is asking developers to update their integrations to other Claude models in the interim.
The company apologised for the disruption, acknowledged the abruptness of the situation, and said it intends to share more information within 24 hours. It also made its intentions clear: it believes the Anthropic Claude export control action is based on a misunderstanding and is actively working toward restoring access. Whether that’s realistic depends on how quickly — and whether — the Commerce Department can be persuaded that the specific jailbreak demonstrated doesn’t represent the novel national security risk it’s been treated as.
The Bigger Picture: AI and Government Authority Are on a Collision Course
This episode is going to matter beyond Anthropic. The US government’s ability to invoke export control authority over domestic AI model releases — not just semiconductor shipments or hardware exports, but software and model weights — is an assertion of regulatory power that the industry hasn’t fully reckoned with yet. Export controls have traditionally governed physical goods and specific dual-use technologies. Applying them to a large language model’s API access is a different kind of move, and it sets a precedent that every frontier AI lab should be watching closely. The Anthropic Claude export control case may become the defining test of how far that authority can stretch.
Anthropic’s situation is particularly complicated because Mythos Preview’s security research applications are genuinely dual-use in the traditional sense — the same capability that helps Mozilla patch its software could, in theory, help a hostile actor find vulnerabilities faster. But Anthropic’s counter-argument is credible: if the capability threshold you’re concerned about is already met by GPT-5.5 and other publicly available systems, restricting one company’s model doesn’t meaningfully reduce risk. It just creates an uneven playing field.
The Trump administration’s track record on AI regulation has been mixed — keen to keep the US ahead of China in the race for AI dominance, but also increasingly interventionist when it perceives security risks, however defined. Thursday’s directive suggests that tension isn’t going away. Frontier AI labs operating at the capability levels Anthropic, OpenAI, and Google DeepMind are now reaching will increasingly find that their product launches are national security conversations, whether they want them to be or not. Dario Amodei’s next few days of negotiations with Washington over the Anthropic Claude export control ruling may end up being as consequential as any model release Anthropic has made.
Source: 9to5Mac
Frequently Asked Questions
What triggered the Anthropic Claude export control directive?
According to Axios, a company claimed it successfully jailbroke Claude Mythos, alarming the Trump administration over potential national security risks. The Commerce Department then issued a directive requiring a license for any export, re-export, or domestic transfer of the affected models, prompting Anthropic to disable them entirely.
Which Claude models are still available after the shutdown?
Anthropic confirmed that only Fable 5 and Mythos 5 are affected. All other Claude models remain fully accessible. Users are being defaulted to Opus 4.8 or their previously selected default model while the situation is resolved.
Is Anthropic challenging the US government’s decision?
Anthropic has been careful with its language, but it stated publicly that it believes the directive is based on a misunderstanding — specifically, a narrow and non-universal jailbreak involving capabilities already present in other publicly available models, including OpenAI’s GPT-5.5. The company says it intends to work toward restoring access as soon as possible.
What was Claude Mythos Preview and why does it matter?
Claude Mythos Preview was a highly advanced AI model designed for security research, capable of identifying software bugs and vulnerabilities. Access was limited to vetted partners through Project Glasswing. Mozilla reported resolving hundreds of vulnerabilities with its help, illustrating both the model’s power and the sensitivity around its wider release.
