Tech / AI
White House Froze Anthropic's Mythos at 50 Recipients
A White House official's statement to the Wall Street Journal, not any formal directive, froze Anthropic's Mythos partner list at 50. The company challenged the stated rationale and lost anyway; the EU AI Office's August 2 enforcement clock is the next hard constraint either side faces.
On April 30, an administration official told the Wall Street Journal that Anthropic could not expand Claude Mythos to 120 organizations, freezing the partner list at roughly 50.
Mythos Preview launched in April 2026 under Project Glasswing, Anthropic's initiative to find and patch vulnerabilities in critical software before attackers exploit them. Independent evaluators found the model scored 73% on expert-level capture-the-flag tasks and completed all 32 steps of a simulated corporate network attack without human assistance.
Anthropic named 12 launch partners, including Apple, Google, Microsoft, Cisco, Nvidia, and JPMorgan Chase. Additional organizations maintaining critical infrastructure brought the total to approximately 50.
The NSA also received access, even as the Pentagon had invoked Title 41, Section 4713 to designate Anthropic a supply-chain risk. A separate White House directive ordered a governmentwide phaseout of its technology. A federal judge blocked that directive in March; the Pentagon's technology chief declared the designation "in full force and effect" regardless.
The White House Cap
No executive order, no written directive. Dean Ball, a co-author of the White House AI Action Plan, had backed Anthropic through the Pentagon dispute. He described the outcome as "a licensing regime, informal, highly improvised, but a licensing regime nonetheless" with no legislation and no formal regulatory body behind it.
Administration officials, speaking to the Journal anonymously, cited two grounds: potential misuse and Anthropic's alleged lack of compute to serve 120 entities without degrading federal access. A day before the Journal story, the White House had separately been drafting guidance to let federal agencies use Mythos despite the ban.
Anthropic challenged the compute rationale, pointing to new contracts with Amazon, Google, and Broadcom. New Project Glasswing invitations froze anyway. Operational technology providers told Nextgov they were "annoyed"; "the processes for these firms to be granted access are ongoing."
On May 6, Anthropic tapped all of SpaceX's Colossus 1, more than 220,000 Nvidia GPUs, specifically to ease rate limits on Claude Code. Whether any of that capacity could be reallocated to Mythos is a question the company has not answered publicly.
Brussels Wants In
The European Commission has held "four or five" meetings with Anthropic without resolving access, Commission spokesman Thomas Regnier said May 11. Anthropic did not attend a Parliament committee hearing on AI safety on May 5. OpenAI offered EU institutions access to GPT-5.5-Cyber the same week; the Commission accepted.
The AI Security Institute ran Mythos evaluations months before the public reveal, given access specifically to stress-test the model's offensive capabilities before release. That was an evaluation arrangement, not an operational deployment.
The AISI distinction implies the Commission is asking for a product Anthropic has not yet created: deployed, auditable, regulator-accountable Mythos. Whether the White House cap or the EU AI Office's August 2 enforcement clock moves first will test whether GPAI enforcement reaches beyond the EU's own corporate footprint.
MEPs Stéphanie Yon-Courtin and Sandro Gozi have demanded Anthropic share Mythos with EU institutions; neither has received a commitment. The EU AI Office's enforcement clock starts August 2.