The Pentagon threatened Anthropic with the Defense Production Act - the same Cold War-era law that lets the president force companies to prioritize government contracts. That is about as close to "soft nationalization" as Washington gets without actually seizing assets.
The threat came alongside the Pentagon formally designating Anthropic a "supply chain risk," then turning around and offering OpenAI a similar contract with comparable strings attached. The message to the AI industry is clear: cooperate on military applications, or the government has tools to compel you.
Palantir's CEO put it bluntly - the government will nationalize AI if Silicon Valley appears unwilling to support military needs. Sam Altman took a more measured tone, publicly acknowledging he has "thought about" whether building AGI might work better as a government project, though he called that outcome unlikely given the current trajectory. OpenAI's head of national security partnerships pushed back slightly, stating "we control which models we deploy."
The Manhattan Project Parallel
This dynamic has a direct historical precedent. Scientists who built the atomic bomb initially had significant autonomy, but the moment they tried to set conditions on how their creation was used, the government asserted control. AI CEOs are now in a similar position: they built something the military wants, and their leverage decreases the more essential the technology becomes.
The practical question for the AI industry is not whether nationalization happens in some dramatic overnight seizure. It is whether the government gradually accumulates enough contractual obligations, security designations, and regulatory hooks to functionally control deployment decisions without ever formally taking ownership.
For users of tools like Claude and ChatGPT, this mostly plays out behind the scenes for now. But government contracts shape product roadmaps. If Anthropic and OpenAI increasingly build for military and intelligence use cases, that changes what features get prioritized, what safety trade-offs get made, and ultimately what these tools look like for everyone else.