Same Night, Same Clause: Inside Amodei's First Interview on the Anthropic–Pentagon Standoff被封杀当晚，OpenAI 拿到了相同条款——Amodei 首次专访揭开 Anthropic 与五角大楼冲突内幕

February 27, 2026, was a long day for Anthropic.

That afternoon, Trump posted on Truth Social (his own social platform), calling Anthropic a company run by "leftwing nut jobs" and ordering all federal agencies to "immediately stop" using Anthropic's technology. Almost simultaneously, Defense Secretary Pete Hegseth announced on X that Anthropic would be designated a "national security supply chain risk" — a label that had previously been used only twice, once for Kaspersky Lab, which has ties to the Russian government, and once for a foreign chip supplier. It had never been applied to a US company.

That evening, Anthropic CEO Dario Amodei sat down for an exclusive interview with CBS News — his first appearance on camera since the incident. Over nearly 30 minutes, he discussed how negotiations with the Pentagon broke down, including a 3-day ultimatum, a "compromise" clause full of vague legal language, and why he refused to budge on two red lines. He characterized the government's action as "retaliatory and punitive" and said he would challenge the supply-chain-risk designation in court.

Original video link: http://www.youtube.com/watch?v=MPTNHrq_4LU

Below are the core points from the interview.

Key takeaways

Anthropic agreed to 98%-99% of the Pentagon's use cases; the dispute concerns only two red lines: domestic mass surveillance and fully autonomous weapons. According to Amodei, these restrictions have never actually been triggered in real military operations.

The Pentagon gave Anthropic a 3-day ultimatum, during which the "compromise" language it sent was full of vague qualifiers like "if the Pentagon deems it appropriate," without making any substantive concessions.

The supply-chain-risk designation had never before been applied to a US company — it had only been used for foreign adversaries such as Kaspersky (Russia) and foreign chip suppliers. Amodei called the move "retaliatory and punitive."

As of the interview, Anthropic had not received any formal government notice — all information came from social media posts by Trump and Hegseth. Anthropic plans to challenge the designation in court.

Just hours after Anthropic was cut off, OpenAI announced a deal with the Pentagon, claiming it included the exact same two red-line restrictions. This suggests the dispute may really be about who gets final say on interpretation.

98% of use cases were already open — the dispute is only over that 2%

The interviewer opened bluntly: "Why not open up Anthropic's AI to the US government without restriction?"

Amodei's answer began with a fact that might surprise many viewers: Anthropic has been the most active of all AI companies in cooperating with the US government. He said Anthropic was the first company to deploy models into the military's classified cloud, the first to build custom models for national security purposes, and is now widely deployed across the intelligence community and the military, spanning areas such as cybersecurity and combat support.

[Note: In July 2025, through its partnership with Palantir, Anthropic signed a $200 million contract with the Pentagon, becoming the first company to deploy frontier AI models on the military's classified systems. OpenAI and Google secured Defense Department contracts around the same time, but until then had only been used on unclassified networks.]

Given this, Amodei said Anthropic had no objection to 98% to 99% of the use cases the Pentagon proposed. The only exceptions were two: domestic mass surveillance and fully autonomous weapons.

"As we defend ourselves against our adversaries, we have to do so in ways that defend our democratic values and preserve our democratic values."

Two red lines: technology has outpaced the law, and AI has made previously impossible things possible

Amodei explained the logic behind the two red lines one by one.

The first red line is domestic mass surveillance. He gave a concrete example: private companies collect Americans' location data, personal information, political affiliations, and so on, and the government can legally purchase this data and then analyze it at scale using AI. The key point is that this practice is not illegal under current law. Before the AI era, such analysis was impractical due to technical limitations, so the law never needed to cover this scenario. Amodei's core argument is that technology has advanced faster than the law can keep up — judicial interpretation of the Fourth Amendment and congressional legislation have not caught up.

[Note: This is not merely theoretical. In 2025, reports showed that ICE (US Immigration and Customs Enforcement) had used facial recognition tools like Clearview AI and NEC's Mobile Fortify system to track individuals. ICE's largest contractor, Geo Group, received over $800 million in contracts in 2025 to mine data from commercial databases and conduct on-the-ground surveillance of suspects.]

The second red line is fully autonomous weapons. Amodei specifically emphasized that he was not talking about the semi-autonomous weapons already in use on the Ukrainian battlefield, but weapons systems with no human involvement whatsoever. However, he is not absolutely opposed to the existence of fully autonomous weapons, acknowledging that "our adversaries may have them someday, and perhaps democratic defense will eventually need them too." He raised two practical concerns.

The first is technical reliability. Amodei said anyone who has used AI models knows these systems have a fundamental unpredictability, and "from a purely technical standpoint, we haven't solved that problem yet." He doesn't want to sell a product that isn't reliable enough, and doesn't want it to result in the deaths of friendly personnel or innocent civilians.

The second is the absence of oversight mechanisms. He described a specific scenario: if there were an army of 10 million drones coordinated and controlled by one person or a small group of people, with no human soldiers on the ground deciding "who to fire on," that level of concentrated power would be unprecedented, and existing accountability structures would completely fail. "We need to have a conversation about who gets to push the button, who has the authority to say no — and we haven't had that conversation yet."

Amodei also revealed that Anthropic had proactively proposed collaborating with the Pentagon on developing fully autonomous weapons technology — prototyping it in a sandboxed environment — "but they weren't interested unless they could do whatever they wanted from the very start."

Inside the breakdown of negotiations: a 3-day ultimatum and a superficial compromise

The interviewer brought up the Pentagon's version of events: "The Pentagon tells us they've agreed in principle to these two restrictions and want to reach a deal. Why couldn't a deal be reached?"

Amodei disclosed negotiation details for the first time. The Pentagon gave Anthropic a 3-day ultimatum: either agree to their terms, or be designated a supply chain risk or face forced acquisition under the Defense Production Act.

Several exchanges took place during those 3 days. At one point the Pentagon sent over a version of the language that appeared, on the surface, to meet Anthropic's conditions, but Amodei said that on closer inspection it was riddled with vague phrases like "if the Pentagon deems it appropriate" or "or do anything as required by law." "It didn't actually make any meaningful concession."

This matches information obtained by ABC News. According to reports, Anthropic told ABC News that the Pentagon's latest contract language made "little progress" toward preventing Claude from being used for mass surveillance of Americans or fully autonomous weapons. The so-called "new language" was "paired with legal jargon" that "allowed these safeguards to be freely disregarded."

He cited a tweet from the Pentagon's chief spokesman Sean Parnell, posted the day before the deadline, as evidence — Parnell reiterated the Pentagon's position: "We will only permit all lawful use." This was exactly the same wording they had insisted on from the start, sharply contradicting the claim of having "agreed in principle."

[Note: Sean Parnell is the Pentagon's chief spokesman and a senior adviser to the Secretary of Defense, appointed by the Trump administration. Another key figure during the negotiations was Emil Michael, Under Secretary of Defense for Research and Engineering (a former Uber executive), who publicly called Amodei a "liar" on social media and accused him of having a "God complex," trying to "personally control the US military."]

Responding to Trump's "selfishness" accusation: a promise to help with the transition even if cut off

The interviewer read out Trump's post that day on Truth Social: "Their selfishness is putting American lives at risk, putting our military at risk, putting our national security at risk."

[Note: The tone of Trump's full post that day was even more heated — he called Anthropic "leftwing nut jobs," accused it of trying to "force the War Department to comply with their terms of service rather than the Constitution," and ordered all federal agencies to "immediately stop" using Anthropic's technology.]

Facing this accusation, Amodei didn't strike back with rhetoric — instead he emphasized a specific commitment: even after these unprecedented measures had been taken, Anthropic was still willing to do everything it could to support the Pentagon — providing technical support until the military completed its transition to another supplier.

The interviewer pressed further: "Are you prepared to pull out?"

Amodei said he was deeply concerned about service disruption. He mentioned having spoken with uniformed military personnel on the front lines, who told him that losing Claude would set them back 6 to 12 months, or even longer. "That's why we tried so hard to reach a deal. But the 3-day ultimatum, the threat of the supply-chain-risk designation — the entire timeline was driven by the Pentagon, not by us."

The boundaries of power in the AI era: who has the right to draw the lines on military AI use?

The interviewer got to the most fundamental question in this dispute: "Why do you think Anthropic — a private company — has more authority than the Pentagon to decide how AI is used in the military?"

Amodei first clarified a fact: as far as he knows, these restrictions have never actually been triggered in real front-line use. This is the 1% of use cases, and there is currently no evidence that anyone actually needs to do these things.

Then he made a key concession — "I actually agree with you." He believes that in the long run, this really should be Congress's responsibility. AI now allows the government to analyze Americans' location data, personal information, and political affiliations at scale, and this isn't currently illegal — which shows the law needs to be updated.

But Congress isn't the fastest-moving institution in the world. Until legislation catches up, Amodei believes that those who best understand the frontier of the technology have a responsibility to temporarily serve as gatekeepers. He framed the essence of this debate with two questions:

"No one wants to be spied on by the US government. No one wants to be spied on by the US government."

These are fundamental rights — the right not to be surveilled by the government, and the right for soldiers, not machines, to make decisions about war.

The interviewer pressed on: "But in the name of basic principles, why should Americans trust you — a private company CEO — to make these decisions instead of the federal government?"

Amodei's answer had two layers. The first layer is market logic: Anthropic is a private company and has the right to choose what it sells and how. If the government isn't satisfied, it's free to switch to another supplier. He said that if the Pentagon said, "Dario, we have a philosophical disagreement, we're going to go with another model," he would disagree but respect that. That would be the normal way to handle this kind of disagreement.

But the government didn't do that. The second layer is his indictment of the punitive measures: the government not only canceled the Pentagon contract, but extended the punishment beyond the Pentagon to other government departments, attempting to revoke broader government contracts as a form of punishment, and issued the supply-chain-risk designation — meaning any private company with military contracts cannot use Anthropic for the military-contract portion of its work. "It's hard to explain this as anything other than punitive."

[Note: Senior senators on the Senate Armed Services Committee wrote a private letter to Anthropic and the Pentagon before the ultimatum deadline, urging both sides to reconcile. The letter stated that the Department of Defense "has no intention of conducting mass surveillance or using autonomous weapons without human involvement — positions we also support in Congress," while also acknowledging that the question of "lawful use" "requires further effort from all stakeholders."]

Lumped in with Kaspersky and Huawei — "it feels very punitive and unjust"

Amodei said that, as far as he knows, the supply-chain-risk designation had never before been used against a US company. Those previously placed on this list included Kaspersky Lab (a cybersecurity company suspected of ties to the Russian government) and foreign chip suppliers.

"Being lumped in with them, it feels very punitive and inappropriate given the amount that we've done for US national security."

[Note: In 2024, the U.S. Department of Commerce banned Kaspersky Lab from providing antivirus services in the United States, citing the threat posed to its operations by the Russian government's cyberwarfare capabilities. Prior supply-chain risk determinations had all targeted companies affiliated with foreign governments — applying the same mechanism to a U.S. domestic company is indeed unprecedented.]

Boeing doesn't tell the military how to use the planes — why is AI different?

The interviewer offered a pointed analogy: "Boeing builds planes for the U.S. military, but Boeing doesn't tell the military how to use that plane. How is this different?"

Amodei responded on two levels. First, the difference in technological maturity . Airplanes have been around for a long time, and generals have a fairly mature understanding of how they work. But AI is advancing at an unprecedented pace — the compute used in models doubles every four months. "We've never seen a pace of innovation like this."

The interviewer pushed back immediately: "If that pace continues, the U.S. government will never be able to keep up. Isn't your logic self-contradictory then? On one hand you say you want to work with the government to protect national security, and on the other you say the technology is moving too fast for the government to keep pace."

Amodei's response was: Congress only needs to catch up once . The technology moves fast, but the issues requiring congressional intervention are few and very specific — namely domestic mass surveillance and fully autonomous weapons. Once a legislative framework is established, it doesn't need to be repeatedly updated.

He brought the argument to a central point: "Why fixate on the 1% of use cases that conflict with our values, instead of advancing the 99% that serve democratic values and can defend this country? We're even willing to study that last 1%, to see if there's a way to do it in line with our values."

If fully autonomous weapons go wrong: the accountability problem of killing civilians by mistake and an army of 10 million drones

The interviewer pressed further: "Give me one or two examples of what could go wrong."

Amodei described two categories of risk. The first is reliability risk : AI targeting the wrong target, killing civilians, lacking the judgment of a human soldier, leading to friendly fire or innocent casualties. "We don't want to sell something we don't think is reliable enough, something that could get our own people killed or kill innocent people."

The second is oversight risk. He asked the audience to imagine a scenario: an army of 10 million drones, coordinated and controlled by one person or a small handful of people. Human soldiers operate under an entire chain of accountability, built on the premise that each individual exercises common-sense judgment. When you concentrate all decision-making in the hands of a very small number of people, that mechanism collapses.

He reiterated that he isn't saying there should never be such a drone fleet — "maybe someday we'll need it, because our adversaries will have it" — but the accountability mechanism must be discussed first. "Who's pressing the button, who has the authority to say no. I think that's entirely reasonable."

The "woke left-wing company" label: Anthropic has consistently and deliberately stayed politically neutral

The interviewer directly quoted Trump's characterization: "Trump called Anthropic a 'woke left-wing company.' Was this decision ideologically driven?"

Amodei listed a series of collaborative records in rebuttal: he attended an energy event hosted by Trump and Senator McCormick in Pennsylvania, discussing ensuring an adequate power supply for AI models; Anthropic signed the AI health app pledge; when the administration's AI Action Plan was released, Anthropic publicly stated it agreed with "much, even most" of its content.

"Anthropic only speaks up on AI policy, because that's the area where we have expertise. We don't get involved in general political issues. This claim that we're not being fair — we have consistently and deliberately tried to stay fair."

He added: "We can't control how others — even the president — perceive us. What we can control is: staying rational, staying neutral, and holding to our convictions."

Governing by tweet: no formal notice, only social media posts

The interviewer asked whether Anthropic would take legal action. Amodei said: "All we received was a tweet. We haven't received any formal communication. There's no actual supply-chain determination document. The government hasn't taken any actual action. There's only a tweet saying what they claim they're going to do."

A U.S. tech company worth tens of billions of dollars, a decision affecting the entire defense AI ecosystem, was conveyed via a social media post. No formal document, no legal process, no direct communication.

"When we receive some kind of formal action, we'll examine it, understand it, and then challenge it in court."

He also pointed out an important legal detail in Hegseth's tweet: Hegseth claimed that "any company doing business with the military cannot do any business with Anthropic," but the law actually only stipulates that Anthropic cannot be used within the scope of military contracts. There's a huge difference between the two. Amodei said bluntly that Hegseth's tweet was "designed to create uncertainty, designed to create fear, uncertainty and doubt."

Asked whether this constitutes an "abuse of power," Amodei sidestepped the question the first time, saying only "this is unprecedented." When the interviewer pressed again, he gave a clearer characterization: this has never happened before, has never been done to an American company, and it's clear from the government's statements and language that "this is retaliatory and punitive. I don't know what else to call it."

Can a deal be reached — "the red lines won't move"

The interviewer asked him to rate, on a scale of 1 to 10, the likelihood of reaching a deal in the future.

Amodei declined to give a number. He reiterated Anthropic's position: the two red lines have been there from day one and won't change . If a consensus can be reached with the Pentagon, a deal is of course possible. "For the sake of U.S. national security, we continue to hope to make this work. But reaching a deal takes two parties."

If he could speak to the president — "opposing the government is the most American thing there is"

The interviewer gave Amodei a chance to speak directly to the president: "If you could speak to the president right now, what would you say?"

He said: We are patriotic Americans. Everything we do is for this country, to support U.S. national security. We were the first to deploy our models to the military because we believe in this country. We draw red lines because crossing them would violate American values, and we're going to defend American values.

Facing the threat of the supply-chain determination and the Defense Production Act — "an unprecedented intrusion by the government into the private economy" — he said Anthropic exercised "the classic First Amendment right to voice dissent."

Opposing the government is the most American thing in the world, and everything we've done here is an act of patriotism. ("Disagreeing with the government is the most American thing in the world, and we are patriots in everything we have done here. We have stood up for the values of this country.")

Business survival — "not only will we survive, we'll do great"

The interviewer asked the question most on the minds of investors and customers: "Do you think Anthropic can survive as a business?"

Amodei said the actual scope of impact of the supply-chain determination is far smaller than Hegseth's tweet implies. "The purpose of the Defense Secretary's tweet is to create uncertainty, to create fear, to make people think the impact will be far greater than it actually is. But we won't let that tactic succeed. We'll be fine."

[Note: Anthropic's projected 2026 revenue is at least $18 billion, with a valuation of about $380 billion. The canceled Pentagon contract was worth $200 million, a small share of its revenue. But the bigger risk comes from the potential ripple effects of the supply-chain determination — many enterprise customers doing business with the Pentagon may, out of risk aversion, cut back their use of Claude.]

The biggest irony: OpenAI got the same terms on the very same night

That very same night, things took a turn.

Just hours after Anthropic was "blacklisted" for holding firm on its two red lines, OpenAI CEO Sam Altman announced on X that OpenAI had reached an agreement with the Pentagon to deploy its models on classified networks. What matters is the terms Altman claimed. He wrote: "Our two most important safety principles are no domestic mass surveillance and human responsibility for the use of force, including for autonomous weapons systems. The Department of War agreed to these principles, reflected them in law and policy, and we put them in our agreement."

This is exactly the same reasoning for which Anthropic was punished.

Multiple media outlets pointed out this contradiction. CNN reported that "it's currently unclear how OpenAI's agreement differs from what Anthropic wanted." Fortune's analysis was sharper: the key difference may lie in the wording — Anthropic tried to explicitly write these restrictions into the contract, while OpenAI agreed that the Pentagon could use its technology for "any lawful purpose," while Altman "also said these restrictions were written into the agreement." How both can be true at once "is unclear."

Axios's report offered another angle: a source familiar with the matter said the restrictions in OpenAI's agreement "reflect existing U.S. law and Pentagon policy," and the intent was not "to invent a new legal standard." Whereas Anthropic's core concern is precisely that — existing law has already failed to keep pace with AI's capabilities.

Mark Dalton, an analyst at the R Street Institute, pointed out a fundamental contradiction in the Pentagon's position: earlier in the week, they considered Anthropic's technology so important to national security that they were considering invoking the Defense Production Act to compel access to it; and now suddenly they're calling it a "supply-chain risk." Something can't be both an "indispensable national security asset" and a "supply-chain risk" at the same time — unless the real issue is who has the authority to say "no."

Altman added a pointed remark at the end of his post: "We're asking the Department of War to offer the same terms to all AI companies, and we think every company should be willing to accept these terms."

The aftermath of this conflict is far from over.

Anthropic said it will challenge the supply-chain risk determination in court. Over 430 employees of Google and OpenAI signed an open letter in support of Anthropic's position, calling for industry solidarity. Google DeepMind chief scientist Jeff Dean personally spoke out against government mass surveillance. But xAI's Elon Musk sided with the government, claiming "Anthropic hates Western civilization."

This dispute has already gone beyond a single contract or the fate of one company. As Amodei repeatedly said in the interview: when AI's capabilities outrun the law, when algorithms can do things the framers of the Constitution never imagined, who holds the line?

OpenAI got a contract with the same restrictions on the very same night, and even the Pentagon ultimately agreed to both red lines. The real question is: does a company have the right to say "no" at the negotiating table, or must it comply unconditionally.

Amodei chose to say "no." The price was a $200 million contract, a "supply-chain risk" label, and a direct attack by the president on social media. He said Anthropic will be fine. Time will tell.

Full interview video: http://www.youtube.com/watch?v=MPTNHrq_4LU