Written by: Techub News
Recently, negotiations between the U.S. Department of Defense and leading AI company Anthropic have stalled, with the Department threatening to classify Anthropic as a "supply chain risk." Anthropic co-founder and CEO Dario Amodei granted an exclusive interview with CBS News, publicly and in detail addressing the core of this controversy for the first time. In this increasingly heated dialogue, Amodei clarified Anthropic's position: full support for U.S. defense while firmly refusing to compromise in two key areas: "domestic mass surveillance" and "fully autonomous weapons."
Deep Cooperation and Two Non-Negotiable Red Lines
At the beginning of the interview, Dario Amodei first clarified a common misconception: Anthropic is not unwilling to cooperate with the U.S. government; on the contrary, they are "the most actively and deeply collaborating AI company with the U.S. government and military." He cited several "firsts" for the company: the first to deploy models on classified clouds; the first to customize models for national security purposes; and their technology has been widely deployed in the intelligence and military sectors for critical areas such as cybersecurity and operational support.
Amodei emphasized that Anthropic's motivation stems from patriotic beliefs: "We must defend our country against authoritarian rivals like China and Russia." However, he quickly shifted to the core premise of cooperation: while countering external threats, it is essential to uphold America's democratic values. It is based on this principle that Anthropic made it clear to the Department of Defense that they could support 98% to 99% of its use cases, except for two areas where they raised red flags.
The first red line is "domestic mass surveillance." Amodei explained that AI technology has made certain previously impractical (or useless) surveillance methods possible. For example, the government can purchase vast amounts of data (such as location information, personal profiles, political tendencies) collected by private companies and conduct large-scale analysis via AI to build citizen profiles. "This is not actually illegal; it was just useless before the AI era," he pointed out, emphasizing that the pace of technological development has outstripped the law, and domestic mass surveillance is "ahead of the law," posing enormous risks of power abuse.
The second red line is "fully autonomous weapons." Amodei specifically distinguished between "partially autonomous weapons" currently in use in places like Ukraine and "weapons that can fire without human intervention." He admitted that opponents may possess such weapons in the future, and thus the U.S. may eventually need them to defend democracy. However, Anthropic has two major concerns regarding this: one is the reliability of the technology, and the other is accountability and oversight.
"Today's AI systems are far from reliable enough to produce fully autonomous weapons," Amodei said. "Anyone who has dealt with AI models understands that they have an inherent unpredictability that we have yet to solve from a purely technical standpoint." Furthermore, if an army composed of drones or robots operates without human oversight, who decides the targets of attacks? How can accountability be ensured? He believes that a nationwide dialogue on how to oversee such weapons has not yet begun, and thus, Anthropic firmly opposes lifting restrictions at this time.
Negotiation Breakdown: Three-Hour Ultimatum and "Punitive" Measures
When asked why an agreement was not reached despite the Department of Defense's "principled agreement" on these two restrictions, Amodei revealed the rushed and pressured negotiation process. He stated that the Department issued a "three-day ultimatum", demanding that Anthropic either agree to their terms or face being classified as a "supply chain risk" or the implementation of measures like the Defense Production Act.
During this brief negotiation window, there were several back-and-forths. Amodei mentioned that the Department sent what appeared to be a text that met Anthropic's requirements, but it contained vague phrases like "if the Department deems it appropriate" or "acting within the legal limits." "This did not yield any substantive concessions," he commented. Subsequent steps also failed to address Anthropic's core concerns. Amodei reiterated that the company has wanted an agreement from the beginning.
In response to the president's accusations on social media that Anthropic is "selfish" and "putting American lives, our military, and national security at risk," Amodei responded that even after the government took "unprecedented supply chain risk designation measures used only against foreign adversaries," Anthropic remained committed to supporting the Department of Defense to ensure continuity of service before their technology was severed and competitors took over. He expressed deep concerns about service interruptions and cited frontline officers who stated that losing Anthropic's technology "would set us back 6 months, 12 months, or even longer."
Amodei pointed the responsibility for the negotiation deadlock to the Department of Defense: "The entire timeline has been driven by the Department of Defense, not us. We are striving to provide continuity of service, and we are working towards an agreement." He hinted that the positions of both sides are still far apart and emphasized that Anthropic's stance is clear and firm: as long as the two red lines are not crossed, they are willing to offer services to all government departments, including the Department of Defense and the intelligence community.
The Responsibility of Private Companies: Bridging Technology Frontiers and Legal Lags
The host posed a sharp question: why should a private company like Anthropic have a stronger voice in AI military applications than the Pentagon itself? Amodei defended this from several angles.
First, he clarified that, to their knowledge, no use case has yet touched the limits of these two red lines. These restricted use cases account for only 1% of the Department of Defense's demands, and there is no evidence that they have been executed or encountered trouble as a result. Anthropic's technology has been widely deployed in the Department of Defense and other government sectors without raising these issues.
Second, Amodei admitted that in the long run, this should indeed be resolved through democratic discussions, with Congress responsible for follow-up legislation. For example, regarding the possibility of domestic mass surveillance, the law (especially judicial interpretations of the Fourth Amendment) or congressional legislation has not kept pace with technological advancements. "But in the long run, we believe Congress should keep up with technological developments. However, Congress is not the fastest-moving institution in the world." He then pointed out that in the current context of congressional lagging, Anthropic, being at the forefront of technology, has a responsibility to make judgements.
"We expect the Department of Defense... to carefully consider these issues," Amodei said, "but in their failure to do so, we need to examine the technology, scrutinize its capabilities (especially in terms of reliability), assess how it is ahead of the law, and even how it diverges from the original intent of the law." He argued that the right to not be subject to government mass surveillance and the right to have decisions about war made by soldiers rather than machines are fundamental American principles.
When asked why Americans should trust a private company CEO rather than the federal government to make these decisions, Amodei offered two points. First, in a free market system, different companies can provide different products based on varying principles. Anthropic's model is not only constrained by usage terms; it has its own "character," with the ability to reliably perform certain tasks while being unreliable for others. "I believe we are best positioned to judge what our models can reliably do and what they cannot." Second, he reiterated that the Department of Defense could simply choose another vendor without these restrictions, which is a normal way to handle such disagreements.
However, the Department of Defense's actions went beyond normal commercial disputes. Amodei pointed out that the government not only extended the dispute to other government departments but also attempted to punishingly revoke Anthropic's other contracts, and activated the "supply chain risk" designation. This means that any private company with a military contract cannot use Anthropic's technology in its military-related business. "They are intervening in private enterprise behavior, which is hard not to interpret as punitive." Amodei emphasized that, to their knowledge, the supply chain risk designation has never been applied to an American company, previously only used against entities regarded as adversaries like Russia's Kaspersky Lab and Chinese chip suppliers. Classifying Anthropic alongside them, considering the contributions the company has made to U.S. national security, seems "very punitive and inappropriate."
When asked if this implies that Anthropic thinks it understands things better than the Pentagon, Amodei did not directly affirm but returned to the logic of the free market and companies' assessments of their own technologies. However, he reiterated that in the long run, it is unfeasible for private companies and the Pentagon to argue this matter; Congress needs to take action. Anthropic is contemplating how Congress can impose certain safeguards that do not hinder the U.S. ability to defeat adversaries while ensuring that the means of victory align with American values.
Balancing Technological Novelty and Values
The host used Boeing as an analogy: Boeing manufactures planes for the military but does not tell the military how to use those planes. How is Anthropic's approach different? Amodei pointed out the key distinction: the novelty of the technology and its exponential pace of development.
"When a technology has matured and stabilized... a general has a pretty good understanding of how airplanes operate. Airplanes have been around for a long time." He acknowledged that there is innovation also in aerospace, but "the pace cannot be compared to AI." Amodei often discusses that AI is on an exponential growth trajectory, with the computational investment in models doubling every four months, "we have never seen such an innovation pace." If this pace continues, the U.S. government will never catch up.
So, since Anthropic has long advocated cooperating with the U.S. government to ensure national security and technology develops so rapidly that Congress cannot keep up, why turn away at this moment? Amodei countered that they are not turning away but delineating two very narrow but crucial red lines. He reiterated that domestic mass surveillance does not help the U.S. catch up with adversaries; rather, it is an abuse of government power (even if technically legal). Regarding fully autonomous weapons, he indeed worries that the U.S. may need to maintain parity, but the technology is not ready, and there is a lack of discussion on oversight and accountability.
"We need to balance the existential need to defeat adversaries—no one emphasizes this more than I do—but we need to fight in the right way." Amodei made an analogy: "It's like asking if we should commit war crimes if our adversaries commit war crimes. I'm not saying that it equates to war crimes. What I'm saying is that the essence of our values is that we must find a way to win while preserving those values. We cannot just degrade in a race. We must have some principles. And these principles are very few."
He called for attention to the 99% of use cases that align with American values and can defend the country, and even to explore the last 1% of use cases to see if there are ways to implement them that conform to those values. "That is our stance, and I believe it is very reasonable."
Potential Risks and Political Positions
When asked to provide examples of what could go wrong, Amodei elaborated again on "reliability" and "oversight." Issues with reliability could lead to weapons misidentifying targets, causing civilian casualties, and lacking the judgement innate to human soldiers, resulting in friendly fire or harm to innocents. "We do not want to sell things we consider unreliable, nor do we want to sell things that could lead to the deaths of our own people or innocents."
The oversight issue pertains to the accountability chain. Human soldiers' decisions have a whole set of accountability mechanisms, provided that "humans" exercise common sense. Amodei posed the question: "Suppose I have an army of 10 million drones, all coordinated by one person or a small group of people... it's easy to see the accountability problem in this scenario, right? Such concentration of power is unworkable." He emphasized that this does not mean we shouldn't have such drone fleets; perhaps we will need them in the future because adversaries have them, but discussions must be held regarding accountability, who presses the button, and who has veto power.
Regarding former President Trump calling Anthropic a "woke leftist company" and questioning whether this decision was ideologically driven, Amodei denied this. He stated that Anthropic strives to remain neutral, speaking only on AI policy issues where they have expertise, avoiding general political topics, and seeking collaboration on potential common ground. He cited examples of past meetings with former presidents and senators regarding securing AI energy supplies, expressing support. "We cannot control how others—even the president—view us. What we can control is remaining rational, neutral, and sticking to our beliefs."
Future Prospects: Upholding Principles and Legal Challenges
Regarding the possibility of reaching an agreement in the future, Amodei said he does not have a crystal ball, but Anthropic's position is clear: the two red lines have been drawn since day one and will not waver. If the Department of Defense can come to a consensus with them, an agreement may be reached. For the sake of U.S. national security, they still hope to foster cooperation, "but an agreement requires efforts from both sides."
If there were an opportunity to speak with the president at this moment, Amodei would say: "We are patriotic Americans. Everything we do is for this country, to support U.S. national security. We actively deploy models to the military because we believe in this country, in defeating authoritarian adversaries, and in defending America. We draw our red lines because we believe crossing them violates American values, and we want to defend those values. When we are threatened by supply chain designations and the Defense Production Act—an unprecedented government intervention in the private economy—we exercise our classic First Amendment rights to voice our opinions and disagree with the government. Disagreeing with the government is the most American thing in the world. Everything we do is an act of patriotism, and we defend this country's values."
When asked if Anthropic, as a company, can survive this upheaval, Amodei appeared confident. He clarified a "misleading" point in the defense secretary's tweet: the tweet claimed that any company with military contracts cannot do business with Anthropic at all, but the law actually states that within the scope of its military contracts, no company can use Anthropic's technology. "This is a very... much more limited impact." He asserted that not only can Anthropic survive, but it will "do well." He accused the secretary's tweet of aiming to create uncertainty, fear, and doubt, but Anthropic will not let it succeed.
Regarding the critics accusing the Pentagon and the White House of "abuse of power," Amodei did not use that term directly, but reiterated its "unprecedented" nature and pointed out that some of the government's statements and wording clearly indicate this is "retaliatory and punitive." "I don't know what other words to describe it: retaliatory and punitive."
On whether legal action will be taken, Amodei noted that so far, they have only received tweets and have not received any formal supply chain risk designation notifications or actual government actions. "When we receive some formal action, we will examine it, understand it, and challenge it in court."
Finally, regarding whether the method of communicating via tweets reflects the government's ability to handle significant national security issues, Amodei was reluctant to comment on a specific government or individual issue. "We are doing everything we can to support U.S. national security. That is why we are committed to seeking an agreement. If we cannot find an agreement, that is why we are committed to completing the transition smoothly, ensuring our warfighters continue to receive support as they enter conflicts. This is also why we are dedicated to opposing actions we believe do not align with this country's values. This is not about any specific person or government. It's about sticking to the right principles."
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
