Anthropic Co-founder Confirms Trump Administration Mythos Briefing

TL;DR: Anthropic co-founder and Head of Public Benefit Jack Clark confirmed at the Semafor World Economy summit that Anthropic briefed the Trump administration on its restricted Mythos model. The disclosure comes while Anthropic is suing the Department of Defense over a supply-chain risk designation assigned in March.

Briefing Despite Active Litigation

Clark framed the ongoing Pentagon dispute as a “narrow contracting dispute” that the company does not want to let disrupt its national security posture. The underlying disagreement is substantive: Anthropic declined to give the US military unrestricted access for use cases including mass surveillance of Americans and fully autonomous weapons, and the OpenAI deal that ultimately went ahead covered ground Anthropic would not. Clark said the company will keep briefing government on future frontier releases regardless.

The Mythos briefing follows reports last week that Trump officials had encouraged JPMorgan Chase, Goldman Sachs, Citigroup, Bank of America and Morgan Stanley to test the model. Combined with Bank of England Governor Andrew Bailey’s public warning this week that Mythos could “crack the whole cyber risk world open”, the pattern shows two transatlantic governments engaging directly with a privately held preview model outside any formal AI regulatory channel.

The Employment Question Diverges from Amodei

Clark also parted gently from CEO Dario Amodei’s public position on AI-driven unemployment. Amodei has previously warned of Depression-era joblessness if frontier capability grows faster than expected. Clark, who runs Anthropic’s economics team, said current data shows only “some potential weakness in early graduate employment” in select industries. Asked what degrees today’s students should choose, he pointed towards subjects requiring synthesis and analytical thinking across disciplines.

Looking Forward

For UK policymakers tracking the Mythos arc, the Clark briefing signals something important: frontier AI governance in 2026 is happening through private briefings to governments in parallel with — not inside — formal regulatory processes. The UK AI Security Institute’s independent evaluation of Mythos, published this week, is the rare public-facing counterweight. UK firms and departments with US exposure should expect their own procurement and risk conversations to start referencing these private briefings as de facto policy inputs, even where the content remains confidential. Boards should ask their AI vendors which governments they have briefed and what commitments, if any, attach to those briefings.