Trump Forces Anthropic to Shut Down Fable 5 and Mythos 5 Globally
The US Commerce Department invoked national security export controls to force Anthropic to pull its most powerful AI models from all users worldwide, citing a reported jailbreak.
Takeaways
- Full shutdown: The US Commerce Department ordered Anthropic to disable Fable 5 and Mythos 5 globally, citing national security export controls.
- Jailbreak trigger: A third-party company reportedly claimed it could jailbreak Mythos 5; Anthropic says the vulnerability is narrow, minor, and also present in OpenAI's GPT-5.5.
- Unprecedented mechanism: Commerce Secretary Howard Lutnick sent the letter to CEO Dario Amodei invoking export control authorities, a framework typically used for hardware, not software APIs.
- All users affected: Because Anthropic cannot isolate foreign nationals, every customer worldwide loses access to both models; all other Claude models remain live.
- Industry precedent: Anthropic warns that applying this standard across the industry would effectively halt all new frontier model deployments.
- Timeline: The government says the lockdown could lift in a few weeks, pending individually validated export licenses from the Commerce Department.
In an unprecedented move, the Trump administration has ordered Anthropic to immediately shut off access to its two most powerful AI models, Fable 5 and Mythos 5, for every user on the planet. The directive, delivered by Commerce Secretary Howard Lutnick to Anthropic CEO Dario Amodei, invokes national security export control authorities to block any foreign national from accessing the models, whether they are inside or outside the United States. Because Anthropic cannot technically segregate its user base that cleanly, the practical result is a full global shutdown of both models.
Anthropic says it received the letter at 5:21pm ET and is complying, while publicly pushing back hard on the reasoning behind it.
The Jailbreak That Triggered a Shutdown
The government's stated concern is a reported method for bypassing, or "jailbreaking," Fable 5. A jailbreak, in this context, means a technique that tricks a model into ignoring its built-in safety guardrails. An administration official told Axios the Commerce Department decided to take the action after another company claimed it was able to jailbreak Mythos, alarming the administration about possible national security risks.
Anthropic reviewed the reported technique and says the threat is being dramatically overstated. Anthropic's understanding is that the government believes it has become aware of a method of bypassing Fable 5, but when Anthropic reviewed a demonstration of the technique, it found a small number of previously known, minor vulnerabilities that other publicly-available models can also discover. In its statement, Anthropic pointed directly at OpenAI as a comparison: the level of capability displayed is widely available from other models, including OpenAI's GPT-5.5, and is used every day by the defenders who keep systems safe.
Anthropic also drew a clear technical distinction between what was found and what would actually be dangerous. A universal jailbreak is a technique that broadly disables a model's safety filters across all topics. A non-universal jailbreak only works in narrow, specific contexts. Anthropic says it has not received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result, and that the potential jailbreaks disclosed are either entirely benign responses or minor findings that provide no Mythos-specific uplift.
What Fable 5 and Mythos 5 Actually Are
To understand why this matters, you need to know what was just taken offline. Fable 5 is the first publicly available product in the same family as Anthropic's powerful Mythos models, which sent shock waves through the cybersecurity world for their superhuman ability to find and exploit cyber vulnerabilities. Mythos 5 is the same underlying model but with its safety guardrails removed, intended for vetted security teams.
The benchmark numbers are striking:
- On SWE-bench Pro, which measures a model's ability to complete difficult software engineering tasks, Fable 5 and Mythos 5 reach 80.3%, vastly outperforming OpenAI's GPT-5.5, which scored 58.6%.
- On ExploitBench, the model family scores 78.0%, compared with 69.0% for Claude Mythos Preview, 40.0% for Opus 4.8, and 34.0% for GPT-5.5.
- Both models are priced at $10 per million input tokens and $50 per million output tokens, which is less than half the price of Claude Mythos Preview, but still ranks as the most expensive of major AI models available globally.
Anthropic had originally revealed Mythos in April but held it back from the public partly because the company said it was extremely adept at skirting cybersecurity defenses and was too dangerous to release. Just this week, it opted to release the Mythos-class model, Fable 5, even as its capabilities exceed those of every model Anthropic had previously made generally available.
A Fight That Was Already Brewing
This shutdown did not come out of nowhere. The administration tried to get Anthropic to pause releasing the latest models but was unsuccessful, prompting the export control letter. Anthropic now finds itself on a Pentagon blacklist deeming it too dangerous for the government's own use, and in a Commerce Department licensing regime deeming it too dangerous for foreign use.
The administration's action also sits in tension with its own stated AI policy. The Trump administration released an executive order to test the most advanced AI models before deployment, but that order is voluntary and explicitly avoids a mandatory licensing regime, something White House chief AI adviser David Sacks secured to prevent what he considers regulatory overreach. An administration official told Axios that Trump "does not want to hurt the industry and wants innovation to continue."
Anthropic's own safety posture, laid out in its Fable 5 launch documentation, was built around the premise that perfect jailbreak resistance is impossible for any model. Instead, Anthropic adopted a defense in depth strategy: make jailbreaks either very narrow or very expensive to produce, combine that with monitoring, and require 30-day data retention on all Fable and Mythos traffic to detect and respond to attacks quickly. The company says more than 95% of Fable 5 sessions run entirely on Fable 5's own responses, with no fallback, and that internal and external red-teaming efforts found no universal jailbreaks after more than 1,000 hours of testing.
Who Gets Hurt, and Who Doesn't
The immediate blast radius is wide. The directive suspends all access to Fable 5 and Mythos 5 for any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect is that Anthropic must abruptly disable both models for all customers, while access to all other Anthropic models is not affected.
The losers here are clear:
- Enterprise customers who had already integrated Fable 5 or Mythos 5 into production workflows face immediate disruption.
- Cybersecurity teams using Mythos 5 through the restricted Project Glasswing program lose their most capable defensive tool.
- Foreign national Anthropic employees are explicitly blocked from accessing the models, even internally.
- Anthropic itself, which is reportedly confidentially filing for a US IPO at a moment when a forced model shutdown is the last thing it needs on its record.
The winners, at least in the short term, are competitors. Microsoft had already announced that Fable 5 is available in Microsoft Foundry and can be used to power agents in GitHub Copilot and Foundry Agent Service, but that access is now frozen. OpenAI's GPT-5.5 becomes the default frontier option for anyone who needs to keep working.
A Precedent That Could Reshape the Industry
Anthropic's statement goes beyond defending its own models. It makes an explicit argument about what this action means for the entire field: if a narrow, non-universal jailbreak is sufficient grounds to recall a commercial model deployed to hundreds of millions of people, then no frontier model can stay deployed. Every major model in production has been shown to have some form of non-universal jailbreak.
The move marks an escalation in Washington's effort to treat cutting-edge AI systems as national security assets. The mechanism used here, export controls under Commerce Department authority, is the same legal framework used to restrict semiconductor exports to China. Applying it to a software model API is genuinely new territory.
The model needs to remain locked down until the U.S. government's national security apparatus is hardened, an administration official said, adding that could happen in the next few weeks. Per Commerce's letter, a license will be required for the export, re-export or domestic transfer of Anthropic's models. Anthropic will also have to submit an additional application for individually validated licenses, and failure to comply would result in financial and civil penalties.
The second-order question is whether this creates a template. If the government can force an immediate global shutdown of a model based on a reported narrow jailbreak, with no public disclosure of the specific concern and no formal review process, every lab building frontier models now has to price in that regulatory risk. That changes the calculus on what to release, when to release it, and how much to invest in models that could be switched off by a letter from a cabinet secretary. Anthropic said it believes the government should have the ability to block unsafe deployments, but only through a process that is transparent, fair, and grounded in technical facts. Its position is that this action was none of those things.