White House vs. Anthropic: Compute, Control, and Competing Priorities

The White House is pushing back on Anthropic’s plan to significantly expand private-sector access to its Mythos AI model, citing concerns over limited compute capacity and potential impact on government usage.

Key developments:

  • Anthropic proposed expanding access from roughly 50 companies to nearly 120, but U.S. officials warned this could strain compute resources needed for federal operations.
  • A forthcoming White House AI policy memo is expected to promote “multi-vendor AI adoption” across agencies—an attempt to reduce reliance on any single provider while easing tensions with Anthropic.
  • According to Axios, the policy may “allow agencies to get around the supply chain risk designation”, a key issue in the ongoing dispute.
  • Meanwhile, capability convergence is accelerating: GPT-5.5 is reportedly approaching Mythos-level cyber capabilities, with former AI adviser David Sacks stating that “all frontier models will reach this level within six months.”

⚖️ Internal Friction Inside Washington

Despite signs of policy softening, divisions remain:

  • Defense leadership appears more hostile, U.S. Defense Secretary Pete Hegseth sharply criticized Anthropic, calling it “run by an ideological lunatic.”
  • At the same time, the White House appears motivated to secure greater internal access to Mythos, even as it restricts broader commercial rollout.

Why Important?

This is not just a vendor dispute, it’s a signal of deeper structural tensions in AI governance:

  • Compute scarcity is now a policy constraint, not just a technical one
  • Governments want priority access to frontier models, even if that limits private-sector scaling
  • The shift toward multi-vendor strategies suggests reduced trust in single-provider dominance
  • Rapid capability convergence (e.g., GPT-5.5 vs. Mythos) means model advantage may be short-lived

In short, Washington is trying to balance three competing goals: control, competition, and capability access, and it’s not fully aligned internally on how to do it.

Sources

  • https://www.wsj.com/tech/ai/white-house-opposes-anthropics-plan-to-expand-access-to-mythos-model-dc281ab5
  • https://www.bloomberg.com/news/articles/2026-04-30/white-house-ai-memo-hits-issues-driving-anthropic-pentagon-feud
  • https://www.axios.com/2026/04/29/trump-anthropic-pentagon-ai-executive-order-gov