Morning Edition · Monday, June 29, 2026
Washington Reopens Anthropic's Mythos Model to More Than 100 Trusted United States Firms
Two weeks after ordering Anthropic to disable its most capable model for foreign nationals, the Commerce Department cleared access for vetted American institutions.

On June 12, the United States government, citing national-security authorities, ordered Anthropic to suspend all access to its Fable 5 and Mythos 5 models for any foreign national, including Anthropic's own non-citizen employees. Officials said they had learned of a technique to bypass Fable 5's safeguards. Anthropic countered that the technique was narrow, unlocking certain cybersecurity capabilities in one specific case rather than defeating the model's protections universally. The company also argued that recalling a deployed model on that basis would halt new releases across the industry.
Two weeks later the administration reversed much of that order. Commerce Secretary Howard Lutnick wrote to Anthropic that "appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model," clearing the model for more than 100 United States institutions, including Fortune 500 firms and government agencies. Commerce also dropped the requirement for an export license when trusted organizations let their non-citizen employees use Mythos 5. A Russian-language technology channel described the move as the first relaxation since the government asserted control over the distribution of Anthropic's products.
The episode establishes a precedent that matters more than the specific model. The government demonstrated it can enable or disable a commercially deployed frontier model and decide which domestic entities may use it. The list of approved companies has not been published, though many are believed to belong to Anthropic's Project Glasswing program.
- If true, who benefits
The United States government gains a demonstrated lever over which firms may run a frontier model, and Anthropic with its vetted corporate and government partners gains privileged access that competitors lack.
- The nuance
The approved list and the criteria for "trusted partner" status remain unpublished, and Lutnick's letter left the consumer-grade Fable 5 restrictions in place while reserving the right to revoke access at any time.
An open-source-intelligence read of how likely this story is true with its real nuance, not a judgment of any outlet. It assesses the claim, weighing independent and adversarial reporting. How we label confidence.
What this means
The United States has shown it will treat a frontier model's access controls as an instrument of national-security policy, gating not only foreign access but which American firms qualify as trusted. That converts a commercial product into a licensed strategic asset and gives the most capable labs a strong incentive to align closely with the government that can revoke their distribution.
What to watch
- Whether the list of approved institutions or the criteria for "trusted partner" status becomes public, which would reveal how selective the access regime really is.
- How allied governments and foreign enterprises respond, since losing access to Mythos pushes them toward downloadable non-United States models they can run without a license.
Observations to monitor, not financial advice.
Synthesized from: Polylog editors · Semafor · Anthropic
Part of a tracked trend
AI Sovereignty and Export Controls on Frontier Models
Over the next 3-6 months, governments increasingly treat frontier AI models as strategic national assets — extending export controls to model access itself and backing domestic 'champion' labs as sovereignty plays.
More from this edition
- Oracle Puts AI-Driven Job Cuts in a Securities Filing, a First for a Company Its Size
- A Chinese Open-Weight Model Is Reported to Match Claude on Cybersecurity Tests
- xAI Says Grok 4.5 Runs on a 1.5-Trillion-Parameter Foundation, but Only in Private Beta
- United States Data-Center Construction Spending Passes Transportation Infrastructure
- A New Paper Argues Prompt Injection Cannot Be Fully Defended in Today's Architectures
- Researchers Propose Axioms to Test What a Model's Latent Reasoning Actually Represents
- A Training Paradigm Tries to Make LLM Agents Plan Ahead Instead of React
- Google Makes Its AI Coding Team Permanent and Pushes Into Midtraining
- Meta's Segment Anything 3 Pushes Open-Vocabulary Perception Toward Production
- OpenAI Maps Which European Jobs AI Will Change, Grow or Automate
- Anthropic Presses Lawmakers to Rebuild Institutions for an AI 'Exponential'