Anthropic’s Amodei defends open-weight stance following critique from Palantir’s Karp
Dario Amodei says Anthropic has never advocated banning open-weight models but disputes claims that broad AI access necessarily helps defenders more.
The dispute between Anthropic's Dario Amodei and Palantir's Alex Karp is, on its surface, a philosophical argument about open-weight AI models. Underneath, it is a turf war dressed in the language of national security.
Amodei's letter is careful to deny the most inflammatory claim attributed to him — that Anthropic wants open-weight models banned. He does not deny the more useful claim: that Anthropic benefits when open distribution is restricted. The distinction matters. A company that builds frontier closed models has a structural interest in any policy that raises the cost of replicating its work. Amodei is not arguing from principle alone; he is arguing from position.
Karp's counter is sharper than it looks. His warning about Europe is not really about Europe. It is about what happens when incumbents use regulation to slow challengers. The Palantir-Nvidia alliance on open-weight models for U.S. government work is a direct play for the federal AI procurement market, framed as patriotism. Karp is telling Washington that the best American AI stack is one the government can inspect, modify, and deploy without paying frontier-model rents to a handful of labs.
The real leverage point is compute. Amodei's own summary gives it away: keep powerful chips out of authoritarian hands, stop industrial-scale distillation, require safety testing. Each of those measures constrains who can train and distribute capable models. Open-weight policy is downstream of chip policy. Whoever controls the silicon controls the ceiling of every open release.
For the labor market, the signal is indirect but real. Federal AI procurement decisions made in the next twelve months will determine which vendors hire aggressively, which contract out, and which treat open-weight deployment as a service line rather than a product. The companies winning those contracts will need engineers who can operate, harden, and integrate open models inside classified and regulated environments — a different skill profile than the research-heavy hiring that defined the last cycle.