Anthropic quietly pulled its latest AI models, Mythos and Fable, from public access for weeks after a government demand—then restored them just as quietly. The episode marks the most aggressive U.S. intervention yet under new AI export review rules.
Anthropic restored access to its Mythos and Fable AI models on Thursday after receiving formal government clearance. The models—Fable 5, in particular—had been available for only a week before the U.S. Commerce Department's Bureau of Industry and Security (BIS) demanded temporary restrictions, citing concerns over potential national security risks tied to advanced reasoning capabilities.
The move stems from the Biden administration's October 2023 executive order on AI safety, which empowered BIS to review frontier models that could pose risks in areas like cyberattacks or bioweapons design. Anthropic, a leading AI safety company, voluntarily complied with the request, limiting access to Mythos and Fable to research partners and vetted enterprise users.
Mythos is Anthropic's next-generation large language model, optimized for complex reasoning, while Fable 5 is a faster, more efficient variant designed for real-time applications. Both were unveiled in late January to limited beta testers. Within days, BIS flagged Fable 5's ability to autonomously exploit software vulnerabilities as a trigger for review. Anthropic's chief executive Dario Amodei said in a statement that the company worked closely with regulators to demonstrate robust safeguards.
After a six-week review, BIS concluded that Mythos and Fable 5 met the 'responsible release' criteria under the new framework. The restoration allows Anthropic to offer the models through its API and Claude interface again, with added usage monitoring and watermarking. 'We believe this process sets a precedent for how frontier AI developers can earn trust while pushing boundaries,' Amodei noted in a company blog post.
Analysts say the episode signals a hardening stance from Washington. Daniel Castro, director of the Center for Data Innovation, called it 'a turning point in AI governance—the first real test of export controls on model weights rather than chips.' The review process remains opaque, but the outcome suggests companies can still operate in the U.S. if they proactively address safety concerns.
Anthropic now plans to expand Mythos access to academic researchers and international partners, pending similar approvals. Watch for the Commerce Department to release formal guidance on frontier model reviews later this year, which could shape how every major AI lab—from OpenAI to Google DeepMind—navigates future launches.
AI
→ Neutral
Anthropic Restores Access to Mythos and Fable AI After Government OK
Fable 5 was available for a week before the government demanded access limits.
Katelyn Chedraoui
CNET
2 min read
7/10
Washington
Key Takeaways
- Anthropic restored Mythos and Fable 5 AI models after a six-week government review by the U.S. Commerce Department's BIS.
- Fable 5 was publicly available for only seven days before BIS demanded restrictions due to autonomous vulnerability exploitation capabilities.
- The review was the first major test of the Biden administration's October 2023 executive order on AI safety and export controls.
- Mythos is a next-generation large language model for complex reasoning; Fable 5 is a faster variant for real-time applications.
- Restored models include added usage monitoring and watermarking; Anthropic plans to expand access to academic researchers pending approvals.
"We believe this process sets a precedent for how frontier AI developers can earn trust while pushing boundaries. — Dario Amodei, Anthropic CEO"
"It's a turning point in AI governance—the first real test of export controls on model weights rather than chips. — Daniel Castro, Center for Data Innovation director"
Frequently Asked Questions
Mythos is Anthropic's next-generation large language model designed for complex reasoning tasks. Fable 5 is a faster, more efficient variant optimized for real-time applications. Both were developed under Anthropic's safety-first approach and are part of the Claude family.
The Bureau of Industry and Security flagged Fable 5's ability to autonomously exploit software vulnerabilities, which raised national security concerns. Under the Biden administration's AI executive order, frontier models with such capabilities may require review before broad public release.
Fable 5 was publicly available for only one week before restrictions were imposed. The entire review process lasted approximately six weeks, after which Anthropic received government clearance to restore access to both models.
Anthropic added usage monitoring, watermarking, and continued API-level controls. The company also committed to limiting access to vetted enterprise and research partners pending further government guidance.
This episode sets a precedent for U.S. export controls on AI model weights. Other frontier labs like OpenAI and Google DeepMind may face similar reviews. The Commerce Department is expected to release formal guidance later this year.
The Bureau of Industry and Security reviews frontier AI models under the October 2023 executive order. It assesses capabilities in areas like cybersecurity and bioweapons. Companies voluntarily comply or face potential licensing restrictions. The review is opaque, but outcomes depend on demonstrated safeguards.
Topics
Original source
www.cnet.com
Discussion
Join the discussion
Sign in to post a comment or reply.
No comments yet. Be the first to share your thoughts!