What Happened
In a dramatic reversal, the Trump administration — previously “laissez-faire” on AI governance — is now quietly negotiating with China to establish an emergency communication channel for AI, prompted by shared alarm over Anthropic’s Mythos model. The discussions are taking place ahead of President Trump’s state visit to China this week.
Anthropic withheld Mythos from public release after discovering it could autonomously find zero-day vulnerabilities in “every major operating system and every major web browser.” The company launched Project Glasswing, granting limited defensive access to roughly 50 security partners including major tech firms and the Linux Foundation. Mozilla has since confirmed Mythos discovered hundreds of previously unknown flaws through automated fuzzing.
Former National Security Advisor Jake Sullivan revealed he had urged the incoming Trump team to continue Biden-era AI dialogue with China, but the administration initially showed no interest. “That’s all changed in the past few weeks,” Sullivan said.
Sources
📰 LA Times — Fears of an AI breakthrough force the U.S. and China to talk
📰 Bruce Schneier — How Dangerous Is Anthropic’s Mythos AI?
📰 Anthropic — Project Glasswing
Why This Matters
When an AI model is powerful enough to force two geopolitical rivals to the negotiating table, we’ve entered genuinely new territory. Mythos represents the first time an AI capability has been treated not as a product or a research milestone, but as something closer to a weapons-grade technology requiring diplomatic coordination.
The fact that Anthropic chose to withhold rather than monetize its most capable model is significant — and the industry is watching closely to see whether this sets a precedent or remains an exception. Meanwhile, the security implications are staggering: thousands of high-severity vulnerabilities, some undetected for decades, now catalogued by an AI. The race between AI-powered offense and defense just became the defining challenge of cybersecurity.