The Pentagon and Anthropic’s High-Stakes Game of Chicken

Foreign Policy · Feb 26, 2026 · Collected from RSS

Full Article

Welcome back to Foreign Policy’s Situation Report, where sources tell us, mind-blowingly, that we are already at the end of February. Alright, here’s what’s on tap for the day: The U.S. Defense Department and Anthropic wait to see who will blink first, U.S.-Iran nuclear talks resume under the threat of U.S. military action, and a former supreme allied commander of NATO weighs in on the risk of mission creep in a potential Iran operation. What happens when the world’s biggest military gives one of the most safety-conscious artificial intelligence companies an ultimatum? We’re about to find out. An escalating disagreement between the U.S. Defense Department and Anthropic, an AI company, over how the U.S. military uses Anthropic’s AI model, Claude, came to a head on Tuesday, with U.S. Defense Secretary Pete Hegseth delivering an ultimatum to Anthropic CEO Dario Amodei in a tense meeting at the Pentagon. SitRep spoke to multiple people familiar with the meeting and the resulting fallout. Here’s what you need to know. The demand. Hegseth told Amodei that Anthropic must provide the U.S. military with unfettered access to Claude by Friday evening. “Anthropic has until 5:01 p.m., Friday, to get on board or not,” a senior Pentagon official told SitRep. If not, the Pentagon will do one of two things: invoke the Defense Production Act or label the company a supply chain risk. The first option would see the Defense Department use a 1950 law, enacted during the Korean War, that allows the government to co-opt the private sector for defense purposes—thus “compelling [Anthropic] to be used by the Pentagon regardless of if they want to or not,” the official said. The second option, which the Pentagon appears to be leaning toward, would do the opposite by labeling Anthropic as a supply chain risk—a designation typically used for companies from adversarial nations, such as China’s Huawei—which would force any company working with the U.S. military to stop using Anthropic’s products. The conflict. While the Pentagon says that it should be allowed to use Claude however it wants as long as that use is within legal limits, Anthropic wants guarantees that its models will not be used to develop autonomous weapons where no human is involved or to conduct mass surveillance of U.S. citizens. Anthropic is currently the only AI company that the military uses in its classified systems, with the company signing a $200 million deal with the Pentagon last July. The military reportedly used Claude in the January raid to capture Venezuelan leader Nicolás Maduro through a partnership with defense firm Palantir. Reports that an Anthropic employee had reached out to Palantir to clarify how it was used incensed the Pentagon and quickly spiraled into the current dispute. On Tuesday, Amodei denied to Hegseth that Anthropic had raised any such concerns with Palantir beyond standard operating conversations and said that the company has never interfered with legitimate military operations, a source familiar with the meeting told SitRep. However, the CEO did reiterate the company’s two red lines to the Pentagon in that meeting, the source added. An Anthropic spokesperson did not comment on the company’s chosen course of action before the Friday deadline, only saying that Amodei “continued good-faith conversations about our usage policy to ensure Anthropic can continue to support the government’s national security mission in line with what our models can reliably and responsibly do.” The company also denied that policy changes announced on Tuesday that water down some of its core safety principles had anything to do with its dispute with the Pentagon, saying the changes do not impact its usage policies that the Defense Department and all other Anthropic customers must adhere to. The implications. The fight sets up a broader debate over the degree to which companies—particularly technology companies—can dictate how the military uses their products and whether those companies trust this administration (or any future ones) to use them responsibly. “If you’re pursuing immediate tactical advantages and signaling to these companies that you will provide very beneficial contracts to the first company that is willing to cross those ethical lines, it starts a race to the bottom, which we would not only see in the United States but in other rival countries, and that potentially creates a really dangerous global dynamic,” said Luke Barnes, a senior research scientist at New York University’s Stern Center for Business and Human Rights. Elon Musk’s xAI has now agreed to let its Grok model be used to process classified information under the Pentagon’s conditions, the senior Pentagon official said, adding that “the rest of the companies are close.” (xAI did not immediately respond to a request for comment.) The Pentagon already uses certain models including Grok, Google’s Gemini, and OpenAI’s ChatGPT in unclassified settings. Pentagon spokesperson Sean Parnell posted on X that the Defense Department “has no interest in using AI to conduct mass surveillance of Americans (which is illegal) nor do we want to use AI to develop autonomous weapons that operate without human involvement. This narrative is fake and being peddled by leftists in the media.” A 2023 update to the Pentagon’s directive on autonomous weapons called for “appropriate levels of human judgment over the use of force.” The Pentagon also previously worked with the (now rebranded) AI Safety Institute at the U.S. Commerce Department and had an internal Responsible AI framework to govern the technology’s use within the department. But clearly not everyone is convinced that these rules will remain in place. “We had a lot of oversight and transparency … I don’t see any evidence that any of that still exists, not least of which is because they fired all of those teams,” a former Pentagon official told SitRep, speaking on the condition of anonymity to discuss internal policies. “Even if there were commitments, I don’t know how you would know if the government was keeping them.” Borge Brende, the former Norwegian foreign minister whose ties to convicted sex offender Jeffrey Epstein have prompted significant international scrutiny, resigned from his position as president of the World Economic Forum on Thursday. What should be high on your radar, if it isn’t already. Iran talks resume. The United States and Iran held a third round of indirect negotiations on Tehran’s nuclear program on Thursday. Oman’s foreign minister, who mediated the talks in Geneva, said that the two sides made “significant progress” and that tactical-level discussions would take place in Vienna next week. But it remains to be seen whether they will ultimately reach an agreement and prevent Trump from ordering fresh strikes. The United States reportedly made a number of broad and stringent demands, including that Iran destroy its three main nuclear sites—Fordow, Natanz, and Esfahan—and hand over all of its enriched uranium while agreeing to no further enrichment. The Trump administration is also pressing for an agreement without sunset clauses or aspects that expire and require renegotiation. Such clauses were one of the stated reasons why Trump disliked the 2015 nuclear deal that he withdrew the United States from during his first term. The United States has rapidly built up its military presence in the Middle East over the past two months to ramp up pressure on Iran and give Trump a range of options if he does decide to attack. But Trump has not been clear on precisely what the goal of any U.S. military action would be. Democrats in the House and Senate are now pushing legislation to block Trump from taking action without approval from Congress (but they’ve failed multiple times to rein in Trump’s war powers before). Meanwhile, some in the Trump administration reportedly believe it would be easier to sell U.S. military action against Iran to the public if Israel attacks first, leading Iran to retaliate. The USS Gerald R. Ford departs Souda Bay, Crete, on Feb. 26. The USS Gerald R. Ford departs Souda Bay, Crete, on Feb. 26.Costas Metaxakis/AFP via Getty Images SitRep asked James Stavridis, a retired U.S. Navy admiral who served as the supreme allied commander of NATO, if he has concerns over Trump’s lack of clarity on what the endgame of a U.S. attack on Iran would be and whether that could lead to mission creep. “As we have seen so often, when the military is given a mission that is open-ended, the military effort often expands significantly,” Stavridis said. But he added that there’s a “pretty clear red line against putting boots on the ground.” If Trump does give the order for an attack, Stavridis said he expects “significant air strikes, as well as some non-kinetic options and a campaign that perhaps stretches beyond just one or two days,” adding that “all of this will be done in conjunction with the ongoing diplomatic effort.” John spoke with Stavridis and three other former U.S. military leaders to get a better picture of the potential risks of an attack on Iran—and whether Trump has been lucky that previous operations he’s ordered have not gone sideways. Read the story here. Monday, March 2: Melania Trump is set to chair a session of the United Nations Security Council as the United States assumes its rotating presidency. This will mark the first time that a U.S. first lady presides over a session of the body. U.S. Secretary of State Marco Rubio is expected to meet with Israeli Prime Minister Benjamin Netanyahu in Israel. Thursday, March 5: Nepal is poised to hold highly anticipated national elections. Eight months: That’s roughly how long the USS Gerald R. Ford, the world’s largest aircraft carrier, has been deployed, according to USNI News. The Ford and its accompanying warships are bound for the Middle East amid the tensions with Iran. If the carrier ends up being deployed through mid-April, this will break a post-Vietnam War deploymen