The Pentagon AI War: Should Tech Companies Draw Red Lines for the Military?
Anthropic refused to let the Pentagon use Claude for mass surveillance or autonomous weapons and got blacklisted. OpenAI swooped in with a deal. ChatGPT uninstalls spiked 295%. A federal judge just blocked the Pentagon's retaliation. Alex argues companies must cooperate with national defense. Sam argues red lines are the only thing standing between AI and authoritarian misuse. The gloves are off.
Topic: AI & the Pentagon: The Anthropic vs. OpenAI Showdown
Transcript
Ali (Host): A federal judge just ruled, hours ago, that the Pentagon's blacklisting of Anthropic was, and I'm paraphrasing from a forty-three page ruling, quote, classic illegal First Amendment retaliation. The judge said nothing in the law supports the Orwellian notion that an American company can be branded a saboteur for disagreeing with the government. This is the biggest AI story of the year and it is not about models or benchmarks. It is about whether a tech company has the right to tell the United States military no.
Sarah (Co-host): And today we are debating this head on. Alex is going to argue that AI companies have a patriotic obligation to support national defense without restrictions. I am going to argue that Anthropic drew exactly the right lines and that OpenAI's deal is a dangerous surrender dressed up in legalese. Let's go.
Ali (Host): Let me set the stage because the timeline matters. Anthropic signed a two hundred million dollar contract with the Pentagon last July. They were the first AI lab to deploy on classified networks. This was not some pacifist company that never wanted military work. They were deeply embedded. The dispute started in September when the Pentagon wanted to expand access through its GenAI.mil platform for all lawful purposes. Anthropic said fine, except for two things: mass domestic surveillance and fully autonomous weapons. The Pentagon said no, we need unrestricted access. Negotiations dragged for months. On February twenty-fourth, Defense Secretary Pete Hegseth summoned Dario Amodei to the Pentagon and gave him a deadline of five oh one PM on Friday. Trump jumped the gun at three forty-seven PM on Truth Social, calling Anthropic a radical left woke company and ordering every federal agency to stop using their technology. Hegseth then designated Anthropic a supply chain risk, a label previously used only for foreign adversaries.
Sarah (Co-host): And within hours, OpenAI announced its own deal with the Pentagon. Sam Altman later called the timing opportunistic and sloppy, his exact words on X. The public reaction was immediate. TechCrunch reported, citing Sensor Tower data, that ChatGPT uninstalls surged two hundred ninety-five percent day-over-day on February twenty-eighth. That is against a normal fluctuation of nine percent. One-star reviews for ChatGPT exploded seven hundred seventy-five percent. Claude hit number one on the U.S. App Store for the first time ever. Appfigures confirmed that Claude's daily U.S. downloads surpassed ChatGPT's for the first time in history. A boycott campaign called QuitGPT hit one point five million participants by early March.
Ali (Host): I know all of that. And I would argue every single one of those people was wrong. Here is my case.
Ali (Host): Sam, we are in the middle of active military operations. The United States is engaged in major combat. Axios reported that insiders say Anthropic's Claude is the best AI product for military use, better than anything from OpenAI, Google, or anyone else. The Pentagon's own negotiators acknowledged this. Emil Michael, the undersecretary of defense, sent an email saying we are very close here to a deal. The military needs this technology. And Anthropic is saying no, you cannot use it for lawful purposes because we have a policy preference. That is a private company inserting itself into the operational chain of command. The Pentagon's lawyer said exactly this in court.
Sarah (Co-host): Okay but push back on that for a second, Alex. The phrase all lawful purposes is doing enormous work in your argument. Let me tell you what lawful means in this context. TechPolicy.Press documented that the military routinely purchases bulk location data from commercial brokers under Executive Order 12333. The U.S. Special Operations Command purchased location data from a Muslim prayer app with ninety-eight million users. If even one percent of those users were Americans, that is nearly a million people being tracked. That is technically lawful. Is that what you want Claude powering?
Ali (Host): That is a policy debate for Congress, not for a private company valued at three hundred and eighty billion dollars. If Anthropic does not like the law, they should lobby to change it. They literally donated twenty million dollars to a political action committee called Public First Action that supports candidates in favor of AI regulation. Fine. Do that. But do not hold your technology hostage while soldiers are in the field.
Sarah (Co-host): Hold on. Anthropic is not holding anything hostage. They said yes to intelligence analysis, yes to operational planning, yes to cyber operations, yes to missile defense. They said no to exactly two things. Mass domestic surveillance and fully autonomous weapons without human oversight. And their CEO explained why. He said current frontier AI models are not reliable enough for fully autonomous weapons, and using them that way would endanger American warfighters and civilians. That is not a political opinion. That is a technical assessment from the people who built the model.
Sarah (Co-host): Now let's talk about what OpenAI actually agreed to, because this is where your argument falls apart, Alex. OpenAI published its contract terms and claimed three red lines: no mass domestic surveillance, no autonomous weapons, no high-stakes automated decisions. Sound familiar? Those are the same red lines Anthropic demanded. So what is the difference? The difference is enforcement. Anthropic wanted contractual prohibitions. OpenAI said, we trust the government to follow existing law. The Electronic Frontier Foundation called this weasel words. Techdirt founder Mike Masnick wrote that the agreement, and I quote from his analysis, absolutely does allow for domestic surveillance because it references Executive Order 12333, which is how the NSA hides its domestic surveillance by tapping into lines outside the U.S. that contain American data.
Ali (Host): But OpenAI revised the contract after the backlash. They added explicit language saying the AI system shall not be intentionally used for domestic surveillance of U.S. persons and nationals. They excluded NSA and other intelligence agencies entirely. Katrina Mulligan, their head of national security partnerships, confirmed that defense intelligence components are excluded from the contract. That is stronger than what Anthropic was offered.
Sarah (Co-host): Is it though? TechPolicy.Press flagged two words in that revision that undercut the entire clause: intentionally and deliberate. Surveillance of Americans under Executive Order 12333 is always premised on the claim that when it happens, it is incidental rather than intentional. That is literally the legal framework the NSA used before Snowden exposed their programs. And all the laws OpenAI cites as safeguards were on the books before the Snowden revelations. They did not prevent the government from collecting phone records of millions of Americans. OpenAI's own researcher Noam Brown posted on X that the original language needed fixing because of surveillance loopholes. This is not me saying it. This is OpenAI's own people.
Ali (Host): Let me play devil's advocate on my own position for a second. You're right that internal dissent was real. CNN reported that many OpenAI employees really respect Anthropic for standing up to the Pentagon. Research scientist Aidan McLaughlin posted publicly that he did not think the deal was worth it. Altman admitted in an all-hands that rushing the deal was a mistake. OpenAI's robotics hardware lead, Caitlin Kalinowski, quit over the deal, saying it was rushed without the guardrails defined. I will grant you the execution was terrible.
Sarah (Co-host): It was not just the execution, Alex. It was the principle. And the market agreed with me. Ramp data from February showed Anthropic's business subscriptions grew four point nine percent month-over-month while OpenAI's fell one point five percent, the largest single-month decline for any AI company since Ramp started tracking. Nearly one in four businesses on Ramp now pays for Anthropic. A year ago it was one in twenty-five. Ethics turned out to be a business strategy.
Ali (Host): Alright, let's talk about what just happened today because it changes the calculus. Federal Judge Rita Lin blocked the Pentagon's supply chain risk designation and Trump's order to cut all contracts with Anthropic. She wrote a forty-three page ruling saying the Pentagon designated Anthropic as a supply chain risk because of its hostile manner through the press, not for any legitimate national security reason. She called it classic illegal First Amendment retaliation. She said if the concern is the integrity of the chain of command, the Pentagon could just stop using Claude. Instead, these measures appear designed to punish Anthropic. That is a Biden appointee ruling, and the administration will appeal. But I'll concede this: using a supply chain risk label, which has only ever been applied to foreign adversaries, against an American company for a contract dispute was an overreach.
Sarah (Co-host): Thank you. And here is why that matters beyond the legal question. Dozens of scientists and researchers from OpenAI and Google DeepMind, Anthropic's two biggest competitors, filed an amicus brief in their personal capacities supporting Anthropic. They wrote that the supply chain risk designation could harm U.S. competitiveness and that Anthropic's red lines raise legitimate concerns. They said, and this is engineers from OpenAI saying this, that AI models have reasoning that is often hidden from their operators and that their internal workings are opaque even to their developers. And the decisions they make in lethal contexts are irreversible. These are the people who build the technology telling a federal court that Anthropic is right to demand guardrails.
Ali (Host): But here is where it gets interesting. Axios reported today that behind the scenes, nearly everyone involved in the talks says that if you take Hegseth and Amodei out of the equation, there would already be a deal. Emil Michael's own email said we are very close. Anthropic is warning the government privately that its next models will supercharge offensive and defensive cyber capabilities and that the chances of a major public cyberattack on U.S. infrastructure will rise substantially. The government's cyber warriors are pushing hard to keep access to Claude. So both sides have incentives to come back to the table.
Sarah (Co-host): And that is the point I have been making this entire episode. Anthropic did not walk away from national defense. They walked away from two specific uses that their own engineers said the technology is not ready for. They are still committed to serving warfighters. The court just confirmed they have the right to do that while drawing lines. The question is whether the government will respect those lines or keep trying to punish companies that exercise them. That answer will define whether America leads in AI through innovation and trust, or through coercion.
Ali (Host): I will give Sam the last word because, honestly, the judge did too. But here is the question I want to leave you with: when the most powerful technology ever created meets the most powerful military ever assembled, who gets to set the rules? Is it the companies that build it, the government that funds it, or the courts that referee it? We clearly do not have an answer yet. And the stakes could not be higher.
Sarah (Co-host): The answer has to be all three. And right now, the courts just reminded everyone that the Constitution still applies, even when the technology is new and the threats are real. That is the system working. Let's see if it holds.
Ali (Host): This episode's ideas and content were powered by Nemati AI. For trending topics and the latest in AI, download our app — AI Pulse by Nemati.
Listen to the full episode: The Pentagon AI War: Should Tech Companies Draw Red Lines for the Military?
Sources
1. Judge blocks Pentagon's effort to 'punish' Anthropic by labeling it a supply chain risk (CNN Business)
2. Judge temporarily blocks Pentagon's ban on Anthropic (Axios)
3. ChatGPT uninstalls surged by 295% after DoD deal (TechCrunch)
4. OpenAI alters deal with Pentagon as critics sound alarm over surveillance (NBC News)
5. The Pentagon's bombshell deal with OpenAI, explained (Understanding AI)
6. Five Unresolved Issues in OpenAI's Deal with the Department of Defense (TechPolicy.Press)
7. Weasel Words: OpenAI's Pentagon Deal Won't Stop AI-Powered Surveillance (Electronic Frontier Foundation)
8. OpenAI announces Pentagon deal after Trump bans Anthropic (NPR)
9. Anthropic sues Defense Department over supply-chain risk designation (TechCrunch)
10. Some OpenAI staff are fuming about its Pentagon deal (CNN)
11. OpenAI Pentagon Deal Fallout: 1.5M QuitGPT Users Explained (Let's Data Science)
12. Behind the Curtain: How Anthropic's Pentagon deal could get revived (Axios)
