- Anthropic said access to Claude Fable 5 and Mythos 5 has been restored following the lifting of U.S. export controls
- The company announced the update on Wednesday across its Claude platforms and cloud services
- The 2026 Intel Summit will convene leaders to discuss AI, intelligence operations and cybersecurity
Anthropic has announced that access to Claude Fable 5 and Claude Mythos 5 artificial intelligence models has been restored following the lifting of U.S. export controls by the Department of Commerce, marking the redeployment of the models across its platforms and cloud services.
As intelligence and defense communities increasingly integrate advanced AI into mission workflows, discussions around safeguards, data governance and cybersecurity oversight remain central to deployment strategies. The Potomac Officers Club will host the 2026 Intel Summit on Sept. 24. The event will feature panel discussions on agentic AI, data governance, counterintelligence, open-source intelligence and cybersecurity in support of mission operations. Sign up now to join experts shaping the future of intelligence technology.
In an article published Wednesday, the company said access is now available across Claude platforms, including Claude.ai, Claude Code and Claude Cowork, with restoration efforts also underway across Amazon Web Services, Google Cloud and Microsoft Foundry.
The update came days after Anthropic received U.S. government approval to restore access to Mythos 5 for a limited group of U.S. organizations that operate and defend critical infrastructure.
What Did the Commerce Department Say About Fable 5?
According to CNN, Commerce Secretary Howard Lutnick said on social media that export controls on Claude Fable 5, the general-purpose version of Anthropic’s Mythos model, were lifted following coordination with Anthropic and U.S. government stakeholders.
“Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America’s leadership in AI,” Lutnick said in a post on X Tuesday.
What Are Anthropic’s Safety Classifiers?
Anthropic said it addressed prior concerns by deploying updated safety classifiers designed to detect and block potentially harmful cybersecurity-related prompts, including jailbreak techniques identified in external reports.
The company said the system blocks the targeted behavior in the vast majority of cases and redirects some blocked requests to earlier model versions with stronger safeguards.
What Is the Shared Industry Framework for AI Jailbreaks?
Anthropic said it is working with Amazon, Google, Microsoft and other Glasswing partners on a proposed shared framework for assessing AI jailbreak severity. The framework evaluates jailbreaks across four criteria:
- Capability gain
- Breadth of capability
- Ease of weaponization
- Discoverability
Anthropic also said it is launching a HackerOne program, allowing external security researchers to submit potential jailbreak findings for review.
What Has Led to the Export Control Directive?
The Verge reported that the export control directive was imposed after concerns emerged regarding potential jailbreak techniques affecting Mythos 5 and Fable 5. The restrictions, introduced in June, temporarily limited access for foreign nationals, including enterprise users and some company employees, while government reviews were conducted.
Anthropic said access was initially restricted because it could not reliably verify user nationality in real time at the point of enforcement.
The directive was later lifted following weeks of negotiations between Anthropic and the U.S. government.
What Are Anthropic’s AI Security Collaboration Commitments to Government?
Anthropic outlined four commitments to expand collaboration with U.S. government partners:
- Pre-release access and evaluation — Expanded early access for government partners to test models and safeguards before deployment, supported by Anthropic technical staff
- Rapid information sharing — Fast notification of significant jailbreaks or misuse patterns, along with sharing of updated safeguards and threat intelligence
- Joint research support — Dedicated teams, compute resources and technical expertise to advance government-led AI security research and evaluation
- Common industry bar — Collaboration with government and industry peers to develop shared standards for frontier AI security and evaluation
