Author: Xiong Lei
Edited by: Xu Qingyang
On June 12, local time in the United States, a rare confrontation between government and enterprises in the history of AI regulation suddenly erupted this Friday.
Citing national security concerns, the U.S. government issued an export control order to Anthropic , demanding an immediate suspension of access to its Fable 5 and Mythos 5 AI models by all foreign entities. This order has a very broad scope, affecting not only foreign users outside the U.S., but also foreign citizens within the U.S., and even foreign employees of Anthropic.
As a compliance measure, Anthropic had to completely shut down access to these two models for all users—this is currently the only feasible solution to ensure compliance, as the company cannot technically distinguish between "foreign entities" and "U.S. citizens" user groups. Access to the company's other models remains unaffected; users will be automatically reverted to Claude Opus 4.8.
This abrupt halt came completely unexpectedly. The Fable 5 and Mythos 5 were officially released on June 9th, just three days ago. The sudden removal of both models has caused widespread shock in the tech world and the AI community.
01 What exactly are the two models?
To understand the core tensions of this controversy, it is necessary to first understand what kind of models Fable 5 and Mythos 5 are, and why they have been under the regulatory spotlight from the very beginning.
Mythos is a new model family from Anthropic, surpassing the Opus series in capability level, representing the highest level of capability that Anthropic can currently publicly deploy. The first Mythos-like model, Claude Mythos Preview, was released in April of this year through "Project Glasswing," with access strictly limited to a small number of partners, because its capabilities in the cybersecurity field are too powerful to be widely available.
Fable 5 is the first Mythos-level model officially released to the public. Its capabilities surpass all previously available models released by Anthropic, achieving top-tier performance in almost all benchmarks, including software engineering, knowledge work, visual understanding, and scientific research.
To enable public release, Anthropic has equipped Fable 5 with a dedicated security mechanism—in high-risk areas such as cybersecurity, biology, and chemistry, the model will automatically block responses and fall back to Claude Opus 4.8 processing.
Mythos 5 is a version built on the same underlying model but with less security protection. It is only available to approved organizations that have previously obtained access to Project Glasswing and is positioned as a professional tool for cybersecurity defenders and critical infrastructure operators. Both models are priced the same: $10 per million input tokens and $50 per million output tokens.
02 The trigger for the command
According to reports, U.S. Commerce Secretary Howard Lutnick sent a letter to Anthropic CEO Dario Amodei on June 12, announcing that Mythos 5 and Fable 5 would be subject to export controls. The immediate trigger for this decision was another company's claim of being able to "jailbreak" Mythos, raising concerns among the Trump administration about potential national security risks.
It is understood that the Trump administration had previously tried to prevent Anthropic from releasing the two models, but failed – which prompted the administration to subsequently take the tougher measure of export controls.
Faced with this sudden order, Anthropic, while complying with regulations, issued a lengthy statement with unusually strong wording, systematically refuting the government's reasoning.
Anthropic believes that the "jailbreak" evidence held by the government only involves a very narrow and non-general attack method, which essentially requires the model to read a specific codebase and fix software vulnerabilities within it—similar capabilities that also exist in other publicly available models, including OpenAI GPT-5.5, and are used daily by cybersecurity personnel for normal system maintenance.
In its statement, Anthropic explicitly stated that using "the existence of a limited number of potential jailbreaks" as the standard for recalling deployed business models would effectively bring all new deployments of cutting-edge models across the industry to a complete standstill. The company also emphasized that the government's action did not follow the legally mandated procedures of transparency, fairness, and based on technical facts, as previously publicly called for.
Anthropic stated that it will comply with the government's legitimate instructions and is actively communicating with the government to restore access as soon as possible. More technical details will be released within the next 24 hours. The company believes the incident stemmed from a misunderstanding and sincerely apologizes for any inconvenience caused to users.
The following is the full text of the statement issued by Anthropic entitled "Regarding the U.S. Government's Order to Suspend Access to Fable 5 and Mythos 5":
Based on national security authorization, the U.S. government issued an export control directive requiring the suspension of access to Fable 5 and Mythos 5 for all foreign nationals, regardless of their location within or outside the United States, including Anthropic's foreign employees. The practical effect of this directive is that we must immediately shut down Fable 5 and Mythos 5 access for all users to ensure compliance. Other Anthropic models are unaffected.
We received the instruction at 5:21 p.m. ET today. The letter did not specify any national security concerns. To our understanding, the government believes it has mastered a method to bypass Fable 5's security measures, a so-called "jailbreak" technique. We reviewed a demonstration of this technique and found that it could only identify a small number of previously known, minor vulnerabilities. These vulnerabilities are relatively simple, and other publicly available models can detect the same issues without jailbreaking.
In a blog post published by Fable, Anthropic outlined our stance on its security measures, as follows:
We have established a robust security mechanism that significantly reduces the likelihood of Fable being misused for cybersecurity and related tasks. In fact, some users have commented that our protection measures are overly stringent.
In the weeks leading up to Fable's release, Anthropic collaborated with the US government, the UK AI Security Institute, numerous third-party organizations, and its internal team to conduct thousands of hours of red team testing on Fable's security. The test results demonstrated that Fable's security was significantly superior to any previously deployed model.
Currently, no testers have been able to find a "universal jailbreak" method—that is, a jailbreak method that can completely bypass model security protections and unlock large-scale network attack capabilities.
We believe that no model provider can currently achieve perfect jailbreak protection. All industry-standard protection mechanisms are susceptible to being breached by "non-universal jailbreaks" (i.e., obtaining partial network information under specific circumstances), and universal jailbreak methods may emerge in the future. We clearly stated this point when Fable 5 was released.
Given that perfect jailbreak protection is currently unfeasible, Anthropic has adopted a defense-in-depth strategy for Fable 5. Our goal is to make jailbreak attacks either extremely narrow in scope (against non-general jailbreaks) or extremely costly (against general jailbreaks), coupled with comprehensive monitoring to quickly detect and contain any successful attacks. This is also why Anthropic requires customer data to be retained for 30 days—although this policy has a real impact on our customer relationships, it helps us research and address jailbreak risks.
We adhere to this defense-in-depth strategy. It effectively reduces the risks associated with Fable, bringing them to a level comparable to existing deployed models in the industry.
To date, we have not even received any formal disclosures regarding non-general jailbreaks that could have harmful consequences. The potential jailbreak cases disclosed to us have either been completely harmless or merely minor discoveries that do not demonstrate Mythos's proprietary capabilities.
Currently, the US government has only provided us with verbal evidence regarding a potential, limited-scope, non-general-purpose jailbreak method that essentially requires the model to read a specific codebase and fix software vulnerabilities within it. To our knowledge, this potential jailbreak method has been shared with the government. We have reviewed a report—which we believe forms the basis for the government's directive—and verified that the level of capability described in the report is also prevalent in other models (including OpenAI's GPT-5.5) and is used daily by cybersecurity defenders to maintain system security. We will release more details within the next 24 hours.
We will comply with the government's legitimate directive to disable access to Fable 5 and Mythos 5 for all users. However, we do not agree that discovering a limited-scope potential jailbreak method is sufficient grounds to recall a commercial model already deployed to hundreds of millions of users. Applying this standard to the entire industry would effectively bring all leading model providers to a complete standstill in new model deployments.
We have publicly stated that governments should have the right to block the deployment of insecure models through transparent, fair, clear, and technically based legal procedures. This action did not follow these principles.
We sincerely apologize for any inconvenience caused to our users. We believe this incident stemmed from a misunderstanding and are actively working to restore access as soon as possible.
