Anthropic's Fable 5 released globally after three weeks of US export restrictions

Anthropic's advanced AI model Fable 5 has been made globally accessible from 1 July 2026, after the US government lifted previously imposed export controls. Commerce Secretary Howard Lutnick withdrew the licence requirement for both Fable 5 and Mythos 5 on 30 June, more than three weeks after access to both models was abruptly frozen.

Mythos 5, the more powerful of the two models, remains available on a limited basis for the time being. Anthropic is releasing that model to a select group of US organisations and is gradually expanding access through the so-called Glasswing programme, in coordination with the US government.

The restrictions began on 12 June 2026, three days after the launch of both models on 9 June. The US government intervened on national security grounds and temporarily required Anthropic to obtain export control licences for use outside the US.

What led to the restrictions

Anthropic launched Fable 5 and Mythos 5 on 9 June 2026. Both models share the same underlying training weights but are configured for different applications. Fable 5 is intended for general use with extensive safety filters. Mythos 5 has fewer such restrictions and is aimed at defensive cybersecurity applications.

Shortly after the launch, Anthropic itself warned the US government that the models could pose risks to national security. That warning signal appears to have been a contributing factor in the government's decision on 12 June to freeze global access. It was an unusually swift government intervention that drew attention across the broader AI sector.

On 26 June, the US government gave Anthropic permission to make Mythos 5 available again to a limited group of trusted US organisations, as an interim step towards broader release.

Fable 5 broadly available, Mythos 5 via selective programme

With the licence requirement lifted on 30 June, Anthropic was able to open Fable 5 to the world a day later. The model is available through Amazon Web Services, Google Cloud and Microsoft Foundry.

A different approach applies to Mythos 5. Because the model has fewer safety filters and is aimed at sensitive cybersecurity applications, Anthropic is releasing it to fewer organisations. This is being done through Project Glasswing, a programme with more than forty participating companies. Glasswing members include Microsoft, Apple, Google, Amazon Web Services, the Linux Foundation, Cisco, Nvidia and Broadcom. Anthropic is working with the government to gradually expand access to broader groups of domestic and international partners.

About Anthropic

Anthropic was founded in 2021 by former OpenAI employees, including CEO Dario Amodei and President Daniela Amodei. The company is headquartered in San Francisco and has a European office in London.

The funding the company has raised is substantial. Amazon has invested a total of $8 billion across multiple tranches. Google has invested more than $2 billion. In March 2025, Anthropic raised $3.5 billion in a Series E round, at a post-money valuation of $61.5 billion. Alphabet is one of the largest shareholders with approximately 14 percent.

Despite its broad investor base and commercial scale, Anthropic positions itself as a safety-focused AI company. That positioning played a direct role in the events surrounding the export controls: it was the company itself that alerted the government to the risks posed by its own models.

What this means for European users and the broader AI sector

For European companies and developers who wanted to use Fable 5, the 1 July release marks the end of a nearly three-week interruption. Access to Mythos 5 via Glasswing remains limited to selected partners for now; when international organisations outside the US will be eligible has not yet been announced.

The incident illustrates how export controls on AI models have become a concrete operational risk for companies that depend on access to US models. Where export restrictions have traditionally applied to hardware such as chips, cloud-based models have now directly affected accessibility for users outside the US. For European policymakers working on their own AI regulation and strategic autonomy, this is an example of how quickly external decisions can affect the availability of AI infrastructure.