Full Breakdown
NSA Loses Access to Anthropic’s Advanced AI Models After Export Controls
6/24/2026, 4:09:20 AM
Core Event: Export Controls Shut Down Mythos 5 and Fable 5
In June 2026 Trump administration issued an export-control order that forced Anthropic, led by CEO Dario Amodei, to block foreign access to its AI models, Mythos 5 and Fable 5. Lacking real-time nationality verification, Anthropic disabled both models, ending NSA testing under Director Gen. Joshua Rudd.
Background & Context: Rapid Shift in U.S. AI Policy
The export order followed a June 2 executive order urging AI labs to share new models with government and a rollback of prior AI safeguards. Administration also approved chip sales to China, marking a swift reversal in U.S. AI strategy.
Timeline of Key Actions
- June 2, 2026 – Executive order requests voluntary model sharing.
- June 11, 2026 – Sen. Warner says Gen. Rudd reported Mythos breached NSA systems “in hours.”
- June 12, 2026 – Export-control directive forces Anthropic to restrict Mythos 5 and Fable 5.
- June 22, 2026 – Media report details shutdown.
Data & Statistics: Reach and Findings
Project Glasswing gave 200 partners—Amazon, Apple, Google, Microsoft, Nvidia, JPMorgan, Linux Foundation, and Five Eyes partners (Australia, United Kingdom, Canada, New Zealand)—access to Mythos Preview. The model uncovered many flaws, including a 27-year-old OpenBSD vulnerability. Over 100 security experts signed a protest letter.
Official Statements & Responses
President Trump told Axios the firm was a security risk and said he was unsure if Defense Production Act was required. Anthropic’s spokesperson thanked administration for “ongoing partnership” and pledged infrastructure protection. Warner cited Gen. Rudd’s claim that Mythos breached classified systems within hours. Security experts warned ban will push open-source or Chinese models.
Criticism & Opposition
Security researchers say the ban hampers red-team testing and vulnerability discovery, while adversaries can still use open-source or foreign models. European policymakers warn dependence on U.S. frontier AI threatens the continent’s AI sovereignty despite its AI Act, NIS2, and Cyber Resilience Act.
Conflicting Reports & Gaps
Sen. Warner’s testimony that Mythos breached NSA and Cyber Command systems remains unverified; no independent assessment has confirmed the intrusion’s scope. The exact number of systems affected, specific vulnerabilities discovered, and operational impact on NSA defenses are unknown.
Verbatim Quotes
- “broke into almost all of our classified systems, not in weeks, but in hours.” — Senator Mark Warner
- “Well, not now, but a week ago, maybe,” — Donald Trump
- “But I’m not sure I have to do that.” — Donald Trump
- “Three words triggered the export ban The three words “fix this code” proved sufficient to trigger the export ban.” — Trump administration
Why It Matters: Strategic Implications
Anthropic’s models show AI can locate software flaws at scale, turning vulnerability discovery into a strategic asset. Limiting defender access may shift advantage to adversaries using unregulated tools, while U.S. leadership in frontier AI could erode if allies lose timely access.
What’s Next
Anthropic is adding identity-verification for its Claude suite and exploring compliance pathways for export rules. Congressional oversight of AI-related export controls is expected to continue, and allied nations are watching the policy for possible replication.
