2 min read AI-generated

Jack Clark at Semafor: Anthropic Talks With Trump Admin and Broadens Mythos Access

Copy article as Markdown

Anthropic co-founder Jack Clark dropped two notable lines at a Semafor event: talks with the Trump administration about the next model — and wider Mythos access.

Featured image for "Jack Clark at Semafor: Anthropic Talks With Trump Admin and Broadens Mythos Access"

Jack Clark, Anthropic’s co-founder and policy lead, was surprisingly candid about the company’s relationship with the US government at a Semafor event in Washington on Monday. Two things he said stand out — and both land differently once you know what’s happened over the past weeks.

Talking with the Trump admin — despite the Pentagon dispute

According to Clark, Anthropic is currently in conversations with the Trump administration about its next frontier model. That’s striking because in February the Pentagon had effectively cut off business with Anthropic through cloud resellers and labeled it a supply chain risk. The fight was about guardrails for military AI — Anthropic wanted tight limits, the DoD found them too restrictive.

Clark put it this way: “We care deeply about national security.” And he made clear that government officials need transparency about the technology. Anthropic will keep those conversations going for current and future models. Which agencies are involved, he didn’t say.

Mythos access is being broadened

The second line matters more for developers and enterprise customers: Anthropic is in the process of broadening access to Claude Mythos. So far Mythos Preview has been rolled out only through Project Glasswing and to a handful of launch partners — Apple, AWS, Google, Microsoft, JPMorganChase, CrowdStrike, Palo Alto Networks and a few more.

Broader access doesn’t mean a public release. Mythos’s cybersecurity capabilities are still considered too dangerous for a wide launch — the model has surfaced thousands of high-severity vulnerabilities across operating systems and browsers in internal testing. The real question is: which additional enterprise customers get gated access, and under what conditions?

My take

The two statements fit together. Anthropic talking to the government while staggering access to Mythos isn’t a contradiction — it’s a playbook: show regulators what the model can do before it goes out. After the Pentagon mess, that’s politically smart. Anthropic is learning that pure refusal doesn’t scale if you want to win US infrastructure deals long-term.

For Cowork and Claude Code users, Mythos remains out of reach for now. But the direction is clear: the model is coming — with one foot on the brake.


Sources: