Suspension of Access to Anthropic's Fable 5 and Mythos 5 Models (2026)

Suspension of Access to Fable 5 and Mythos 5 is the decision by US-based artificial intelligence company Anthropic to halt access to its Fable 5 and Mythos 5 models on 13 June 2026, in accordance with an export control directive issued by the US federal government on national security grounds.

The directive prohibits all foreign nationals, including Anthropic’s foreign employees, from accessing the models both inside and outside the United States. In compliance with the directive, Anthropic has closed access to Fable 5 and Mythos 5 for all customers. Other Anthropic models, including Claude Opus 4.8, have not been affected by this measure.

^{Logo of the artificial intelligence company Anthropic - (Anadolu Ajansı)}

Anthropic publicly introduced its latest artificial intelligence model, Claude Fable 5, in a statement dated 9 June 2026. The company described the model as a new capability layer, termed Mythos-class, that surpasses its previous Opus-class models. The announcement stated that the model’s capabilities exceed those of all previous models made publicly available by the company.

Mythos 5 is the full, non-public version of the model, designed exclusively for use by government agencies and selected corporate partners to enhance their systems.

Fable 5 is a publicly available version based on Mythos technology, with additional security measures applied. Anthropic, recognizing the model’s exceptional capabilities, initially granted limited preview and testing access in April only to a small number of organizations, including certain US government agencies.

The Mythos model has been noted for its particular success in identifying software vulnerabilities that had remained undiscovered for decades. It is reported that US authorities and selected companies have used this capability to patch security flaws.^【1】 Experts have warned that the model could accelerate sophisticated cyberattacks, posing risks especially in sectors such as banking that rely on complex and largely legacy systems.^【2】 Tests by the United Kingdom’s Artificial Intelligence Safety Institute (UK AISI) found that the model could bypass defenses and systems in 73 percent of cases.^【3】

In its 9 June announcement, Anthropic acknowledged the risks of releasing such a powerful model and stated that it had implemented security safeguards to prevent misuse in areas such as cybersecurity.^【4】 The company disclosed that, in the weeks leading up to the launch, Fable’s security measures were subjected to over a thousand hours of “red-team” testing with the US government, UK AISI, numerous independent third-party organizations, and internal teams.^【5】

Anthropic stated that these tests demonstrated Fable’s security measures were significantly more effective than those of any previously released model, but that no tester succeeded in finding a comprehensive “universal jailbreak.”

The US federal government, citing national security concerns, issued an export control directive requiring the suspension of access to Fable 5 and Mythos 5 by all foreign nationals, both within and outside the United States, including Anthropic’s foreign employees.

According to Anthropic’s statement, the directive reached the company on 13 June 2026 at 17:21 Eastern Time, and the letter provided no detailed justification for the national security concerns.

It was confirmed that the directive applies not only to individuals outside the United States but also to foreign nationals located within the United States. A US official confirmed that the Department of Commerce issued an export control directive blocking foreign nationals’ access to Fable 5 and Mythos 5.

The Wall Street Journal reported that Commerce Secretary Howard Lutnick sent a letter imposing the restrictions to Anthropic’s Chief Executive Officer Dario Amodei. A government official speaking to American CNBC confirmed that the directive was issued by Lutnick and that its text was prepared with input from officials at the Department of Commerce’s Bureau of Industry and Security.

According to Axios, the directive requires Anthropic to obtain a license for the export, re-export, or domestic transfer of the models.

Anthropic stated that to comply with the directive, it was forced to rapidly disable access to both models for all customers. The company expressed this in its statement as follows: “The clear consequence of this order is that we had to abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Anthropic models will remain unaffected.”

Amazon’s cloud division, AWS, confirmed that Anthropic requested it to cancel access to the models “for all users across all regions”. Users reported being unable to access the models on Saturday.

Although the letter provided no details, Anthropic stated that the government believes it is aware of a method that can bypass Fable 5’s security measures. These security measures were designed to restrict users’ access to Mythos’s core cybersecurity capabilities.

The company stated that it reviewed a study demonstrating that this method had been used to identify certain previously known vulnerabilities. Anthropic noted that these same vulnerabilities could also be detected by other publicly available AI models.

Anthropic said the government provided it with verbal information regarding a narrowly scoped method of bypassing its security measures, based on examining a specific codebase of the model to identify and fix software flaws.

The company reviewed the report it believes underpinned the government’s directive and noted that the capabilities described in the report are also present in other AI models. Anthropic complied with the directive by disabling access to the models but formally objected to the decision. In its statement, the company argued that the findings cited were insufficient justification for blocking public access to a commercially deployed model.

Anthropic also emphasized that security reviews of AI model deployment must be conducted within a transparent, open, and technically grounded process. The statement concluded by noting that efforts are underway to restore access and that apologies have been extended to customers.

The suspension of access to Fable 5 and Mythos 5 follows a series of disputes between Anthropic and the Trump administration in 2026. The conflict began when Anthropic refused to permit the US military to use its Claude model in fully autonomous weapon systems and domestic surveillance operations.

US Defense Secretary Pete Hegseth issued an ultimatum to Anthropic on 24 February, warning that failure to allow the Pentagon to use the company’s AI as required by 27 February would risk termination of government contracts.

US President Donald Trump issued a directive on 27 February 2026 ordering federal agencies to halt the use of Anthropic’s technology. On 6 March, the Pentagon classified Anthropic as a “supply chain risk”.

On 9 March, Anthropic filed a lawsuit against the US government and challenged the Pentagon’s decision in federal court. Subsequently, a US judge ruled that the Pentagon’s directive could not be enforced, thereby allowing government agencies and military-affiliated organizations to continue using Anthropic’s services while the case proceeded.

The Mythos model also emerged in discussions surrounding the presidential executive order on advanced AI models. The order requires companies to voluntarily share with the government any new models deemed to possess advanced cyber capabilities 30 days before granting access to other parties.

A number of individuals, including security personnel from Nvidia, Zoom, and Mercedes-Benz, as well as former US government and Google security officials, sent an open letter to Lutnick urging him to lift the restrictions on Fable 5 and Mythos 5. The letter called for a commitment that future AI risk assessments be conducted through an open, scientific, and transparent process, warning that removing the most capable models without sound justification is dangerous.

Following the decision, debates over AI sovereignty have emerged in various countries. UK Minister for AI and Online Safety Kanishka Narayan stated that the development underscores the need for his country to increase investment in its AI sector. European Commission spokesperson Thomas Regnier noted that the situation highlights “Europe’s need for technological sovereignty” and that the matter is under review.

It has been reported that Anthropic’s leadership held talks with the US government in Washington, DC, regarding the access restrictions. Attendees included Commerce Secretary Howard Lutnick, National Cyber Director Sean Cairncross, Anthropic co-founder Tom Brown, and the company’s head of public policy Sarah Heck.

The U.S. government has granted Anthropic authorization on Friday, June 26, 2026 to make Mythos 5 available to a select group of trusted U.S. organizations.

According to Al Jazeera, this authorization covers the provision of models to U.S. organizations that “operate and defend critical infrastructure.” Designed to identify cyber vulnerabilities, Mythos 5 was previously made available to a broader group of companies under the Glasswing project.

Anthropic announced on Tuesday, June 30, 2026 that the U.S. Department of Commerce had lifted all export controls on Fable 5 and Mythos 5. In a social media post regarding the matter, the company stated, “We have been informed that the U.S. Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5. We will begin restoring access tomorrow and will share an update shortly. We thank our users and everyone who worked with us during the process of returning these models to service for their patience.”

According to Anthropic’s announcement, Fable 5 will become globally available to users on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork starting Wednesday, July 1, 2026 . The company also stated it will reactivate access on AWS, Google Cloud, and Microsoft Foundry as soon as possible.

Access to Mythos 5 has been reopened for a group of U.S. organizations, and the company is continuing to coordinate with the government to expand access to its broader domestic and international partner group within the Glasswing program.

In a letter to Anthropic seen by the BBC and Reuters, U.S. Commerce Secretary Howard Lutnick noted that the company has addressed the risks involved. Lutnick stated in the letter that Anthropic has committed to proactively identifying and mitigating security risks associated with the models. The letter also records that Anthropic has agreed to collaborate with the government on future versions of the models and to report any malicious activity to the government.

Lutnick emphasized that the decision remains subject to review and that export licensing conditions may be reinstated if circumstances change or if Anthropic fails to meet its commitments. In a social media post, Lutnick stated that over the past two weeks the U.S. government has worked closely with Anthropic to ensure alignment across federal agencies and to strengthen “America’s leadership in artificial intelligence” by analyzing and approving Fable 5.

Anthropic has disclosed that, in close collaboration with the government to address the reported bypass, it has trained an enhanced safety classifier designed to detect and block the behavior identified in the report.

According to the company, the new classifier blocks over 99 percent of requests employing the technique described in Amazon’s report; blocked requests are redirected instead to the Opus 4.8 model. Anthropic noted that the new classifier occasionally flags harmless requests during routine coding and debugging tasks and that it is continuing to refine the measure to reduce false positives.

Researchers from the U.S. Department of Commerce’s Center for Artificial Intelligence Standards and Innovation (CAISI) have tested both the previous and new security measures and have concluded that they are exceptionally robust.

Anthropic has announced that it is working with Amazon, Microsoft, Google, and other Glasswing partners to draft a consensus framework for assessing the severity of AI jailbreaks and determining how developers should respond to them.

The proposed framework rates a jailbreak according to four criteria: capability gain, breadth of capability gain, ease of weaponization, and discoverability. The company also announced the launch of a new HackerOne program through which security researchers can submit potential cyber jailbreaks discovered in Fable 5 for review.

Anthropic has stated that it is deepening its collaboration with the U.S. government and will provide extended early access to its models and security measures for designated government partners . The company explained that its commitments encompass pre-release government access and evaluation, rapid information sharing on security measures, dedicated resources for joint research, and a shared industry standard.