When an AI company says no to the Pentagon
Anthropic, the company behind the Claude model, took a stance that very few tech companies would dare to take: it refused contract terms that would give the Pentagon virtually unlimited freedom to use its artificial intelligence models in military operations. This was not just a business decision. It was a deliberate act of corporate dissent against the largest defense machine on the planet, and the consequences came fast. The U.S. Department of War classified Anthropic as a supply chain risk, something that had never happened to an American tech company of this size. In practice, this means the company started being treated almost like a threat to the national security apparatus, simply for having set ethical boundaries on how its technology could be used.
The timing of this confrontation makes everything even more significant. We are living through a moment when governments around the world are trying to figure out how to exercise control over increasingly powerful AI systems. Anthropic’s decision to draw a red line against unrestricted military use highlights a fundamental tension: who really calls the shots when the most transformative technology of our era meets national security interests? The answer, at least for now, seems to be that the U.S. government is not willing to take no for an answer.
And when OpenAI, Anthropic’s main competitor, stepped into the vacuum and struck a deal with the military — accepting exactly the terms its rival had rejected, including a contract clause that allows the use of AI models for any purpose deemed legal — it became clear that market dynamics can also be used as a tool of political pressure. OpenAI CEO Sam Altman defended the decision in a Q&A session on the X platform, arguing that private companies should not be the ones deciding what is or is not ethical in the most sensitive areas. In his words, it might seem reasonable for OpenAI to define how ChatGPT responds to a controversial question, but he would not want a private company deciding what to do when a nuclear missile is heading toward the United States.
The contract in question was worth approximately $200 million, but the financial value is almost irrelevant compared to what is really at stake. The message from the Pentagon was crystal clear: companies that do not cooperate with defense objectives can be sidelined, replaced, and even labeled as security problems. This sets a dangerous precedent because it transforms military cooperation from a business choice into an implicit obligation for any company developing technology considered strategic. And artificial intelligence, without a doubt, sits at the very top of that list.
What the Pentagon actually did and why it matters
To understand the severity of what happened, you need to look closely at the tool the Department of War used as retaliation. The supply chain risk designation was originally designed to be applied to technologies that could help a foreign adversary sabotage critical defense systems. Never before had this classification been used against an American company, let alone as a form of punishment for disagreeing on contract terms.
In practice, this designation means that any company doing business with the Pentagon would be banned from maintaining commercial relationships with Anthropic. As Dean Ball, an AI policy expert who briefly worked on the Trump administration’s AI action plan, pointed out, canceling the $200 million contract would have been well within the government’s rights. But going far beyond that and labeling the company as a supply chain risk amounts, in his words, to attempted corporate assassination. That is because Anthropic depends on sales to major Fortune 500 companies that also serve the Pentagon, as well as cloud computing infrastructure and venture capital.
Several legal scholars have already signaled that this interpretation would probably not survive a legal challenge. Still, the reputational damage and the chilling effect on other AI companies are real and immediate. In the wake of the designation, agencies such as the Department of the Treasury, the State Department, and the Department of Health and Human Services announced they would end their use of Anthropic’s Claude models, migrating to solutions from OpenAI, Google, and in some cases, xAI.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The specter of nationalization and AI control
This episode reignited a debate that many people preferred to keep on the back burner: the possibility of nationalizing AI companies considered strategic to the United States. The idea might seem extreme for a country that has historically championed the free market above almost everything, but the reality is that AI is increasingly being treated as a national security asset, on the same level as nuclear technology or space capability.
Sam Altman himself acknowledged this possibility. In his statements on the X platform, he said it has seemed to him for a long time that maybe it would be better if building artificial general intelligence (AGI) were a government project, though he noted this does not seem very likely on the current trajectory. It is a revealing admission coming from the CEO of one of the largest AI companies in the world.
In practice, the Pentagon’s approach already resembles nationalization by other means. One of the options the Department of War reportedly considered was using the Defense Production Act, a Cold War-era law, to compel Anthropic to deliver an AI model on the military’s preferred terms — a kind of soft nationalization of the company’s productive capacity. Combined with the supply chain risk designation, designed in part to intimidate other AI companies into accepting the Pentagon’s contract terms, the strategy appears clearly adjacent to nationalization.
The problem with this path is that it could destroy exactly what makes American AI innovation so powerful. Tech companies thrive in environments where they can experiment, fail, iterate, and make independent decisions about the direction of their products. Government control exercised directly, especially in military contexts, tends to prioritize short-term defense applications at the expense of fundamental research and safety development for the models themselves. Anthropic, it is worth remembering, was founded specifically with the mission of developing AI in a safer and more responsible way. Forcing a company with that profile to abandon its founding principles to meet military demands is not just an ethical issue — it is a strategy that could compromise American technological leadership in the long run, because it drives away the best researchers and engineers who are attracted precisely to companies that take safety seriously.
On top of that, the international landscape makes everything more complex. If the United States begins treating its AI companies as extensions of the military apparatus, other countries will do the same. China already operates with a much tighter integration between the private sector and the state in AI development. Europe, for its part, has been trying its own regulatory path with the AI Act. How this dispute is resolved will send a clear signal to every democracy still crafting its AI governance policies.
Corporate dissent and the silence of Congress
Perhaps the most revealing aspect of this entire situation is the role of dissent — or rather, the lack of institutional space for it. Anthropic tried to exercise something that, in theory, should be perfectly legitimate in a democracy: disagreeing with terms proposed by a client, even if that client is the federal government. But the disproportionate reaction from the Pentagon showed that, in practice, corporate dissent against military interests comes at an extremely high cost.
The company did not just lose the contract — it earned a label that could damage its business with other government agencies and even with private partners that depend on federal contracts. This kind of retaliation has a ripple effect that goes far beyond Anthropic: it serves as a warning to every other AI company that pushing back against the Pentagon’s demands can have severe and lasting consequences. As Altman said while justifying his deal, a close partnership between governments and the companies building this technology is extremely important. But close partnership and total submission are very different things.
And where is the U.S. Congress in all of this? That is probably the most frustrating question to answer, because the short answer is: essentially absent. In the three years since the launch of ChatGPT, lawmakers have failed to pass any comprehensive federal legislation on artificial intelligence. The Trump administration dismantled the limited AI regulations implemented by the previous administration and has even moved to penalize states that pass their own regulations. This legislative vacuum is what allows the Pentagon to act with such freedom and leaves companies like Anthropic essentially unprotected when they decide to draw the line.
As the original article rightly points out, AI technology advances at the speed of light, but democratic control mechanisms — legislation, parliamentary oversight, elections — move at the speed of a turtle. Without a clear legal framework, every dispute like this gets resolved on the basis of political and economic power, and in that game the Department of War has an overwhelming advantage. The idea of trying to define AI policy through contract negotiations between labs and the government is a poor substitute for real democratic governance, but it may be better than no governance at all.
The question of surveillance and public trust
There is yet another layer that makes this dispute especially sensitive: the track record of governments expanding their surveillance powers through elastic interpretations of existing laws. Over the past several decades, the U.S. executive branch has gradually regained surveillance capabilities it had lost after the Watergate scandals and the Church Committee hearings in the 1970s. Many military activities are shrouded in secrecy, making democratic oversight and accountability extremely difficult.
This constant tendency to push the limits of what the law allows has caused part of the public to lose trust in the government’s intentions. It is no surprise, then, that some people place more faith in a seemingly well-intentioned and brilliant tech executive, like Anthropic CEO Dario Amodei, than in government bureaucrats to define the right policies for AI use.
A recent study published by researchers from ETH Zurich, Anthropic itself, and the MATS program found that AI agents with full internet access can re-identify pseudonymous internet users at scale, analyzing interviews and cross-referencing information with posts on forums like Hacker News, Reddit, and LinkedIn profiles. The AI did this in minutes, while a human investigator would take hours per identification, achieving 90% precision and 68% recall. This kind of capability is exactly what concerned Anthropic during negotiations with the Pentagon: re-identifying anonymous users from publicly available or commercially acquired data was not something possible at scale before, but it may not fall under the classic definition of mass surveillance.
Data centers on the front lines
As if this scenario were not complex enough, the conflict took on an unexpectedly concrete dimension when Amazon Web Services data centers in the United Arab Emirates and Bahrain were struck by Iranian missiles or drones. Although it is not known exactly why the Iranians attacked these facilities, there is speculation that the goal was to disrupt the U.S. military’s use of Anthropic’s Claude models. Despite the Pentagon labeling Anthropic as a risk and declaring it would cease using Claude immediately, reporting from the Wall Street Journal and Axios indicated that the military continued using Claude for target processing in Operation Epic Fury against Iran, on classified networks hosted by AWS.
This raises a question that will become increasingly relevant in future conflicts: as AI becomes essential to military operations, data centers — even those far from the front lines — become legitimate strategic targets. Digital infrastructure is no longer just commercial infrastructure; it is, in practice, combat infrastructure 😬.
What this fight means for the future of artificial intelligence
Regardless of how this specific dispute is resolved in courts or at the negotiating table, the confrontation between Anthropic and the Pentagon has already changed the conversation about artificial intelligence in an irreversible way. Until recently, the AI debate revolved mainly around technical capabilities, existential risks, and market regulation. Now, we are openly discussing the possibility of nationalizing technology, the limits of corporate autonomy in the face of state interests, and the right of a company to simply say no.
Meanwhile, OpenAI closed a funding round of $110 billion that valued the company at $730 billion, with investments from SoftBank, Amazon, and Nvidia. Amazon invested $50 billion, tying part of its commitment to the purchase of Trainium chips and to the achievement of milestones like AGI or an IPO. This move reinforces how the AI ecosystem is rapidly reconfiguring itself around relationships with governments and major cloud infrastructures.
These are questions that will shape not just the tech sector, but the very functioning of democracies in the coming decades. How governments, companies, and civil society negotiate these boundaries will determine whether artificial intelligence is developed with checks and balances or captured by military and political interests with zero transparency.
For anyone following the AI sector, this episode is an important reminder that technology never exists in isolation from power. Language models, computer vision systems, and autonomous agents are incredibly powerful tools, and control over who can use them and how is, ultimately, a political question. Anthropic made a choice that may prove costly in the short term, but could also inspire a broader movement for transparency and accountability in AI development. As Altman himself acknowledged, fostering distrust and conflict between the government and the people building advanced AI systems seems like a terrible idea at a time when this technology threatens unprecedented changes to the economy and society.
What happens from here depends on many factors — public pressure, legislative action, positioning by other companies, and the willingness of governments to accept that dissent is not insubordination, but an essential part of how free societies work 🤔.
