Exclusive Group Retains Access to Claude Mythos Preview Amid White House AI Freeze
WASHINGTON, D.C. — While Anthropic recently suspended public access to its most advanced frontier models following sweeping intervention from the U.S. government, a highly limited cohort of organizations continues to maintain active access to Mythos Preview.
The model variant is an experimental, hyper-potent version of the intelligence engine deployed exclusively for deep cybersecurity vulnerability research.
Inside Project Glasswing's Exclusion Zone
Last week, the White House issued an emergency directive ordering Anthropic to immediately halt the distribution of its advanced AI infrastructure to foreign nationals without explicit, federal licensing. In response, Anthropic pulled its commercial powerhouse, Fable 5, from general availability and frozen the wider rollout of Mythos 5.
However, regulatory enforcement has not fallen uniformly across all user brackets. According to details unmasked by Bloomberg, a tightly insulated network of approximately 200 vetted organizations enrolled in Anthropic's closed cybersecurity research program, Project Glasswing, have quietly retained their access to the highly restricted Mythos Preview environment.
The elite testing collective includes some of the world's largest technological, defense, and financial infrastructure operators:
- Technology Providers: Amazon Web Services (AWS), Cisco
- Financial Institutions: JPMorgan Chase
- Critical Infrastructure Specialists: Industrial cybersecurity giant Dragos
Why Washington Fears "Mythos"
The specific preview model has triggered immense regulatory anxiety due to its unprecedented, autonomous software auditing capabilities. During large-scale benchmark tests conducted earlier this year, Mythos Preview demonstrated a startling capacity to scan, dissect, and identify zero-day vulnerabilities across a massive variety of commercial software suites, web browsers, and foundational computer operating systems.
Because a model that can find unpatched exploits instantly could act as an autonomous cyber-weapon in the wrong hands, the U.S. government has clamped down heavily on its geographic distribution.
| Program Variable | Status Under U.S. Export Order | Impact on Operational Environment |
|---|---|---|
| Claude Fable 5 | Fully Recalled / Blocked | Commercial API endpoints offline globally |
| Claude Mythos 5 | Public Rollout Aborted | Shelved indefinitely for general commercial use |
| Mythos Preview | Restricted Exemption | Active only for roughly 200 approved Glasswing members |
Geopolitical Fractures and the European Exclusion
The selective enforcement of the U.S. export ban has already begun fracturing international research partnerships.
The European Union's cybersecurity agency, ENISA, had previously been invited by Anthropic to collaborate within Project Glasswing. However, sources confirm that ENISA was abruptly notified last week that its access had been fully revoked.
This selective de-authorization confirms that Anthropic is actively auditing its user base on an organization-by-organization basis, prioritizing alignment with aggressive U.S. export controls over international collaboration. The move highlights a major structural transition: high-tier language models are no longer being treated as basic software products, but as dual-use military and strategic defense assets.
Source & References
- Primary Coverage: Early Users of Anthropic Mythos Still Have Access After US Order — Bloomberg
Are regulatory model shutdowns or surprise regional API freezes threatening your application's reliability? Don't let your infrastructure rely on a single vulnerable endpoint. Use our free Cost Simulator to model multi-region fallback strategies, stress-test your token inputs against live alternative rates, and build a resilient multi-vendor roadmap today.