US Orders Suspension of Access to Anthropic's Fable 5 and Mythos 5 Models

What Happened?

On June 13, 2026, the US government issued an unprecedented export control directive ordering Anthropic to immediately suspend access to its artificial intelligence models Fable 5 and Mythos 5 for anyone who is not a US citizen, both domestically and abroad. The order, received at 5:21 p.m. ET, even affects foreign employees of Anthropic, according to the company's official statement. The company announced it must abruptly disable access to these models for all clients—including US citizens—to comply with the regulation, as it is not technically possible to distinguish between domestic and foreign users. Anthropic's other models, including earlier versions like Claude 4, are not affected.

The directive is based on the International Emergency Economic Powers Act (IEEPA) and export control regulations from the Department of Commerce. According to the letter received by Anthropic, the government alleges there is a 'jailbreak' method that bypasses Fable 5's safeguards, exposing national security vulnerabilities. However, Anthropic claims the technique in question simply involves asking the model to review a codebase and find software flaws, a capability available in other public models like OpenAI's GPT-5.5. The company maintains that the identified vulnerabilities are minor and already known to the cybersecurity community, and that models like GPT-5.5 can discover them just as easily without needing a bypass.

Simon Willison, a developer and AI expert, noted on his blog that he still had access to Fable via claude.ai and Claude Code at 9:01 p.m. ET, suggesting the restriction was not implemented immediately across all platforms. The order also requires Anthropic to inform its foreign clients that they must cease using these models, and the company plans to share more technical details within the next 24 hours.

Why Is This Important?

This is the first time the US government has used national security authorities to restrict access to specific AI models, setting a significant precedent that could redefine the regulatory landscape. Historically, AI export controls have focused on hardware, such as the NVIDIA A100 and H100 chips banned for China in 2022 and 2023. However, this directive marks a shift toward controlling AI software, opening the door to future restrictions on language models, vision systems, or generative AI.

The move comes amid growing geopolitical tension around AI. In 2025, the Biden administration had already proposed rules to limit the export of AI models to certain countries, but none were as drastic as this. The government's lack of transparency—providing only verbal evidence of a jailbreak, according to Anthropic—contrasts with the company's stance, which reviewed the report and concluded that the demonstrated capability is widely available. This raises doubts about the proportionality of the measure and its technical basis.

Furthermore, the directive affects Anthropic's foreign employees, which could have implications for talent retention and company operations. Anthropic has offices in the UK and Canada, and its research teams include non-US citizens. The order could force the company to relocate personnel or restructure access to internal models.

Immediate Consequences

• Users and businesses: Clients relying on Fable 5 and Mythos 5 will lose access abruptly, potentially disrupting workflows, development projects, and production applications. Cybersecurity companies that used Fable for code analysis will have to migrate to alternatives like GPT-5.5 or local models.
• Regulatory uncertainty: The action sets a precedent for other AI providers, such as OpenAI, Google, or Meta, to face similar restrictions. The AI community is watching closely, as any model with advanced capabilities could be considered a national security risk.
• Market impact: Stocks of AI companies could experience volatility. The measure may also accelerate the adoption of open-source models, which are harder to restrict, and foster the development of AI capabilities outside the US, especially in China and the EU.
• Anthropic's response: The company plans to share more details within the next 24 hours, but for now there is no clarity on when access will be restored. Anthropic has requested a review of the directive and is evaluating legal options.

Context and Comparisons

This event echoes previous restrictions on AI chip exports to China, but is unprecedented for software models. In October 2022, the US banned the sale of NVIDIA A100 and H100 chips to China, citing national security risks. That measure affected hardware companies, but not software products directly. In 2023, restrictions were expanded to more advanced chips. Now, the government is targeting the software layer, which could have a broader impact since AI models are distributed globally via APIs.

Comparatively, the European Union has adopted a different regulatory approach with the AI Act, which classifies models by risk but does not impose export restrictions. China, meanwhile, has developed its own models like DeepSeek and Baidu ERNIE, which could benefit from this measure if international clients seek non-US alternatives.

The controversy underscores the tension between national security and transparency in AI development. While the government acts on classified information, Anthropic argues the jailbreak technique is trivial and widely available. This raises questions about whether the measure is truly effective or merely shifts the problem to other models. Cybersecurity experts have noted that jailbreaks are a constant challenge, and restricting a specific model does not solve the underlying vulnerability.

What Should Readers Know?

If you are a user of these models, prepare for a temporary disruption. The measure does not affect other Anthropic models or those from other providers, but future directives could expand. The controversy highlights the need for public debate on balancing national security with innovation and transparency. Follow official sources and Anthropic for updates. Meanwhile, consider diversifying your AI tools to mitigate regulatory risks.