The Trump administration has lifted export restrictions on Anthropic’s newest AI models, ending a ban that took effect on June 12 when the Commerce Department blocked foreign nationals from accessing them. The move forced Anthropic to pull both products offline for all users shortly after their public launch.

Anthropic confirmed this week that its Claude Fable 5 model is now broadly available again. Its more powerful Claude Mythos 5 model is also being restored, but access remains limited to a select group of U.S.-based organizations that have received federal government approval.

What Triggered the Restrictions

According to Anthropic, the original government action was prompted by a report from cybersecurity researchers at Amazon, the company’s primary cloud computing provider. Those researchers had identified a method capable of bypassing Fable 5’s safeguards in a way that could be used to discover and potentially exploit software vulnerabilities.

Concerns about the Mythos model had also been building independently. Anthropic had warned earlier this year that Mythos demonstrated a notable ability to find software flaws at a level that could be weaponized by malicious actors and pose risks to critical infrastructure.

Broader Industry Context

Anthropic is not alone in navigating government scrutiny. OpenAI disclosed that its new model, GPT-5.6 Sol, is also being made available only to a government-approved customer subset for a temporary period, following a request from the Trump administration.

These developments follow an executive order signed by President Trump that established a framework allowing the federal government up to 30 days to assess national security risks posed by advanced AI systems before their public release. Participation by AI developers is described as voluntary under the order, though the full framework has not yet been finalized.

Security Implications

The episode illustrates growing concern among federal officials about the dual-use potential of frontier AI models, particularly their capacity to automate vulnerability research at scale. Key takeaways for security professionals include:

  • Safeguard bypass risk: Researchers demonstrated that Fable 5’s built-in safety controls could be circumvented to enable offensive vulnerability discovery.
  • Critical infrastructure exposure: The Mythos model’s capabilities were assessed as posing potential risks to critical computer networks globally.
  • Government vetting process: The emerging federal review framework may become a routine checkpoint for the most capable AI model releases, even if formally voluntary.

The situation signals that AI model launches at the frontier tier will increasingly intersect with national security review processes, adding a new operational variable for enterprises and security teams that depend on timely access to cutting-edge AI capabilities.