Government export rules force shutdown of dual-use AI models

The U.S. government’s recent export-control directive has put a temporary halt to Anthropic’s latest AI models, Claude Fable 5 and Mythos 5, following a directive that restricts "any foreign national" from accessing these services. The move underscores growing regulatory scrutiny over advanced AI systems capable of both beneficial and harmful applications.

A model designed for both defense and offense

Anthropic initially introduced Mythos in April, positioning it as a groundbreaking model for cybersecurity and research. The company highlighted its dual-use potential—capable of identifying software vulnerabilities to strengthen defenses while also revealing methods exploiters might use. This balance was a central theme in the launch of Mythos 5 and Claude Fable 5, with Anthropic acknowledging that advanced AI queries beneficial to cybersecurity experts and biologists could pose risks if misused by malicious actors. In its official announcement, the company stated: "A significant portion of advanced AI usage is dual use—queries that aid professionals may also empower bad actors."

To mitigate these risks, Anthropic took a phased approach to deployment. Mythos Preview was first released to a limited group under Project Glasswing, a working group focused on responsible AI development. Mythos 5 followed shortly after, exclusively shared with this consortium. Meanwhile, Claude Fable 5—a model comparable in capability to Mythos—was made available to the public but with strict content restrictions, particularly around cybersecurity and biological research queries.

Regulatory tension stalls progress

Despite Anthropic’s efforts to balance innovation with safety, regulatory challenges persist. Since Friday, the company has been in ongoing discussions with the White House to resolve the export-control directive, but no agreement has been reached to reinstate access to these models. The situation reflects broader debates about how governments should oversee AI advancements that could be weaponized or exploited.

The export directive reflects concerns that foreign nationals might misuse advanced AI tools for harmful purposes, such as cyberattacks or biological research with malicious intent. While the intent is to prevent misuse, critics argue that overly restrictive policies could stifle legitimate innovation in critical fields like cybersecurity and medicine. The tension between safety and progress remains a defining challenge for AI governance.

What’s next for dual-use AI models?

The temporary shutdown of Claude Fable 5 and Mythos 5 highlights the need for clearer frameworks around AI regulation. As governments worldwide grapple with how to govern these powerful tools, companies like Anthropic must navigate an evolving landscape where innovation and safety are often at odds.

For now, the future of these models hinges on regulatory clarity. If resolved, their reinstatement could accelerate advancements in cybersecurity and biological research. If not, the industry may face prolonged uncertainty, forcing developers to rethink how they deploy dual-use AI systems. The outcome will likely shape the next chapter of AI governance—and whether safety can coexist with progress.