- The White House told Anthropic it opposes expanding Mythos access from ~50 to ~120 companies, citing national security concerns and insufficient computing capacity to safely serve that many users.
- Mythos is capable of finding and exploiting software vulnerabilities at a scale that has rattled government agencies and companies worldwide — Anthropic deemed it too dangerous for general release.
- Anthropic is separately investigating potentially unauthorized access to Mythos, adding urgency to the White House’s security concerns.
- The administration’s block comes despite efforts from both sides to repair a frayed relationship — the White House had previously tried to cut ties over Anthropic’s Pentagon contract dispute.
What Happened?
Anthropic proposed expanding access to Mythos — its most powerful and restricted AI model — from approximately 50 to 120 organizations managing critical infrastructure. The White House said no. Administration officials cited two concerns: security risks posed by a model that can autonomously detect and exploit software vulnerabilities, and worry that Anthropic lacked sufficient computing capacity to serve 70 additional entities without degrading the government’s own access. The rebuff comes as Anthropic separately investigates a potential unauthorized access incident involving Mythos, and as the model itself is already rattling cybersecurity professionals globally with its capability to trigger what some are calling a “bugmageddon” — a flood of exploitable software flaws discovered by AI.
Why It Matters?
Mythos represents a new category of AI risk: a model so capable at offensive cyber operations that its creator restricted its own rollout. The White House’s involvement in managing access — effectively acting as a de facto regulator on a private company’s product distribution — is unprecedented and sets a template for how frontier AI models may be governed going forward. The broader Anthropic-White House relationship remains fraught: the administration has criticized Anthropic for ties to liberal-leaning donors and former Biden officials on its payroll, and two separate court cases involving Anthropic’s Pentagon contract are still active. A former Anthropic researcher was nearly appointed to lead a federal AI evaluation office before the White House abruptly reversed course.
What’s Next?
Anthropic is accelerating deals with Amazon, Google, and Broadcom to expand its computing capacity — a prerequisite for any future Mythos expansion. The company is also weighing an IPO as soon as October and is entertaining funding offers at a valuation exceeding $900 billion, which would make it the world’s most valuable AI startup. The White House has said it’s working to expand government access to Mythos — suggesting the dispute is less about whether the model gets deployed and more about who controls the pace and terms of that deployment. For the broader AI industry, the episode makes clear that the most capable models are now effectively dual-use technologies subject to national security oversight.
Source: The Wall Street Journal
