Hard Fork

At the Pentagon, OpenAI is In and Anthropic Is Out

Kevin Roose & Casey Newton
1 Mar 2026 14 min read 41m
TL;DR

Anthropic refused the Pentagon's demands to drop safeguards against mass domestic surveillance and autonomous weapons, leading to threats of a supply chain risk designation. Within hours, OpenAI signed a deal claiming identical protections—but the language suggests OpenAI may have agreed to looser constraints using legal workarounds, raising questions about whether one company was punished for ideology while the other got preferential treatment.

Key Moments
Dario Amodei (via Kevin Roose)
“These threats do not change our position. We cannot in good conscience accede to their request.”
Anthropic's CEO issued this statement on Thursday as the Pentagon deadline approached, drawing rare praise for invoking conscience in tech policy
▶ 3:29
President Trump (via Kevin Roose)
“The United States of America will never allow a radical left woke company to dictate how our great military fights and wins wars.”
Trump's Truth Social post late Friday escalated tensions by ordering federal agencies to cease using Anthropic's technology, though notably did not invoke the supply chain risk designation
▶ 4:53
Casey Newton
“the Pentagon has said, Well, you know, we're not going to domestically surveil people. That's illegal. Hmm, well, at the same time, Kev, there are other federal agencies right now that have mounted what amounts to a social media dragnet looking through the social media posts of people trying to immigrate to this country”
Newton explains why Anthropic's red line on mass domestic surveillance matters even though the Pentagon claims it won't do it—legal gaps enable functionally equivalent surveillance
▶ 11:10
Kevin Roose
“I have not seen them try to kill a company for what as far as I can tell are contractual disputes and ideological differences.”
Roose contextualizes the severity of the Pentagon's actions against Anthropic compared to historical precedent of US government actions against companies
▶ 17:29
Casey Newton
“My fear is, though, that either through naivete or deception, he has misled us, and we are going to find out sooner or later that in fact those two use cases are not only legal, but they're happening.”
Newton expresses skepticism about whether Sam Altman's safety claims will hold up, noting Altman's history of telling people what they want to hear
▶ 25:04
About the show
›
Hard Fork is a podcast from The New York Times that breaks down the biggest news in tech and AI. Kevin Roose and Casey Newton dive deep into complex stories like Pentagon contracts, corporate governance, and the intersection of technology and government power.
Takeaways
1
The legal/illegal distinction obscures real surveillance Anthropic's refusal to accept an 'all lawful use' clause reveals a crucial gap: mass surveillance of Americans via data brokers is technically legal, so legal language won't prevent it. Without new privacy laws, AI safeguards are cosmetic. The Pentagon can buy data on millions of Americans and feed it to AI without violating any statute.
2
Political vendetta or substantive disagreement? The core mystery is whether OpenAI agreed to materially weaker terms using legalese, or if both companies agreed to identical protections and Anthropic was punished for ideology and personality conflicts. Without seeing the actual contracts, no one—including employees—can verify if these safeguards are real or security theater.
3
Regulatory capture through political pressure OpenAI gained a Pentagon contract by accepting the government's terms; Anthropic was threatened with supply chain risk designation for refusing. This creates a chilling effect: any AI company that wants government work must comply with whatever the current administration demands, regardless of ethical concerns. Employee activism and public pressure may be the only remaining accountability mechanism.