US permits Anthropic’s Mythos AI for vetted organizations

The U.S. Department of Commerce on June 27, 2026 granted Anthropic permission to make its Mythos large‑language model available to vetted U.S. organizations under the agency’s “trusted AI” program [Semafor].

Mythos, a 100‑billion‑parameter model that Anthropic says matches OpenAI’s GPT‑4‑turbo on standard benchmarks, will be accessed via a secure API hosted on Anthropic’s cloud infrastructure [Semafor].

Access is limited to firms that receive a “trusted AI” designation, which requires background checks, export‑control compliance and a safety‑testing commitment before an API key is issued [Semafor].

The rule replaces a prior export restriction that barred models above 10 billion parameters without a license, a rule that kept Mythos in‑house since its internal release in March 2026 [Semafor].

What happened

Anthropic will roll out the Mythos API to the approved firms over the next 90 days, with usage monitored through a reporting portal that logs query volume and content categories.

Why it matters

The “trusted AI” pathway sets a regulatory precedent for U.S. export policy, signaling that compliance will be a prerequisite for market access. It lets defense contractors, critical‑infrastructure operators and regulated fintech firms embed Mythos into mission‑critical pipelines, accelerating the move from research prototypes to production‑grade generative AI. By routing the most capable model through a compliance‑heavy gate, the policy may push non‑U.S. vendors toward open‑source alternatives, reshaping the global AI landscape.

Editor's take

The policy gives Anthropic a privileged distribution channel, potentially cementing a two‑tier AI ecosystem that favors incumbents with regulatory ties over smaller innovators.

Reader poll

Which AI access model should dominate the U.S. market?

• Open access for all developers
• Restricted “trusted” access only
• No AI deployment at all