U.S. President Donald Trump's administration just effectively declared the developer of arguably the United States' most advanced AI models a greater national security threat than DeepSeek or any other Chinese AI developer. After it was clear that Anthropic, the AI lab behind the Claude family of models, would not meet the Department of Defense's Feb. 27 deadline to agree to the Pentagon demands on AI use, Trump announced he was ordering all federal agencies to cease using Anthropic's AI tools, saying that some agencies, like the DoD, would have a six-month phase-out period, describing the company as a "Radical Left AI company" run by "Leftwing nut jobs."
Defense Secretary Pete Hegseth announced he was directing the Department of Defense to designate Anthropic a "Supply-Chain Risk to National Security," meaning that effective immediately, no contractor, supplier or partner doing business with the military could conduct "any commercial activity" with the company. However, Hegseth did not invoke the Defense Production Act, as he had previously threatened, to force Anthropic to work with the military regardless of its terms of service.
The decisions come as a result of a major dispute between Anthropic, whose Claude models were the first ones available to the DoD through its partnership with data platform technology company Palantir, over expanding the military's use of Claude through a $200 million contract signed last year and have the AI company relaxing its terms of use so that the military could use it for "all lawful use."
Anthropic would not budge over two issues. First, over the use of AI for mass domestic surveillance, arguing that powerful AI systems can assemble "scattered, individually innocuous data into a comprehensive picture of any person's life — automatically and at massive scale." Second, over the use of AI for fully autonomous weapons that take humans out of the kill loop entirely, arguing that today's most advanced AI systems are "simply not reliable enough" for such uses.
At its heart, the dispute is unsurprising. Technology companies and the military have frequently disagreed over the use of advanced technology for military purposes for years. It is no surprise that Anthropic is extremely cautious on the matter. The Claude developer, whose Opus 4.6 model sits at or near the top of many aggregate AI benchmarks, was originally founded by employees of ChatGPT developer OpenAI who left the company over concerns that it and its founder, Sam Altman, were not prioritizing safety enough when developing AI models.
By contrast, Anthropic has set and is attempting to maintain a reputation for developing the "safest" frontier AI models. The Pentagon argues that a company should not have a "veto" over its operations. The U.S. Air Force, for example, would never accept conditions from Lockheed Martin on the types of missions it can carry out using the F-35 Lightning II. On the other hand, AI models can be fine-tuned afterwards, generalize across different domains and then carry out actions, making them fundamentally different technologies from fighter jets with clear capabilities.
Anthropic's opposition to the use of fully autonomous vehicles appears to be more technological than ideological. The company offered expanded cooperation on research and development to the Pentagon in order to improve the reliability of fully autonomous weapons and called partially autonomous weapons "vital." The dispute over mass surveillance appears to be not only technological, due to the broad ability for AI to summarize thousands of data points to surveil people, but also on political and ideological grounds under the current administration.
While the U.S. military has some legal authority to conduct domestic surveillance operations, the Trump administration in particular has been quick to deploy federally controlled security forces domestically. Anthropic is almost certainly concerned about how the U.S. government could use its AI models for domestic surveillance under Trump, as it would be a reputational liability for the company.
The U.S. reaction to Anthropic's decision not to meet its demands goes well beyond what would happen if this were just a dispute between the government and a contractor over a particular product. Generally speaking, the contract would be canceled, not awarded or not extended. If implemented, Hegseth's threat to label the company a supply chain risk would be the first time that the DoD has taken such a step against an American company. The label is designed to target companies from adversarial nations on the level of Chinese telecommunications giant Huawei.
Although Anthropic will file a lawsuit against the decision, the designation would require organizations supplying the military to certify that Claude is not used in their supply chains for the specific work that they are doing for the military. Hegseth's statement, however, goes well beyond this, saying that contractors could not have "any commercial activity" with Anthropic.
On the surface, that requirement would mean that companies that supply the DoD and have invested in or sold to Anthropic, such as Amazon, Nvidia and Google, would need to stop and untangle their investments. This goes well beyond the requirements for Chinese AI developers, which largely cannot supply the government, but can still be used by companies that have contracts with the government in some areas and can also buy chips from Nvidia.
Legally, the designation does not go as far as Hegseth's statement. Companies wanting to do business with the Trump administration that have large partnerships with Anthropic may, regardless, seek to reduce those ties over concerns about being passed over in government contracts over ties — even if legal — to the company. Trump's order for all federal agencies to stop using Anthropic's tools, even in non-national security-related applications, also goes well beyond what would be traditionally expected over what is, at heart, a contractual dispute.
The spat between Anthropic and the Trump administration is a part of a broader trend of the White House using heavy-handed tactics against American companies that it views as being inconsistent with the administration's policies and ideology. It harkens back to the Trump administration's executive orders in 2025, effectively sanctioning seven prominent law firms, cutting off their security clearances and pressuring government contractors to halt business with them, over issues including DEI practices being involved in legal cases brought against Trump.
The extent to which the decision is about Anthropic, not the underlying dispute, may even be indicated by the subsequent deal announced by OpenAI after Anthropic's deadline on a similar contract expired. Initially, OpenAI CEO Sam Altman explicitly stated that the agreement with the Pentagon included the company's safety principles on "domestic mass surveillance and human responsibility for the use of force, including for autonomous weapon systems." After backlash, Altman then announced plans to further modify the contract’s terms regarding surveillance, bringing OpenAI closer to Anthropic's position.
From a practical standpoint, the administration's dispute with Anthropic is a net loss for the U.S. military. Blacklisting Anthropic — even with a six-month phase-out period — will create operational challenges as the DoD and Palantir phase out Claude to use other AI models in its stead, like OpenAI's GPT models and xAI's Grok models, which will likely require testing and modifications to perform as well as Claude models do in its workflows. This means that it is possible that, at least initially, the capabilities will be degraded and/or not as advanced as they could have been if they were using Anthropic's models, including in mission-critical areas like visual reasoning, which the military used during its January operation to seize Venezuelan President Nicolas Maduro and in its ongoing military operations against Iran.
Anthropic's models are also clearly currently the best when it comes to writing code, meaning that even on the back end, shifting away from Anthropic to use other models could decrease the military's efficiency in non-combat operations, at least temporarily. The fast nature of AI innovation means that many of Anthropic's competitors will probably have AI models that are as capable or more capable than Anthropic's today within six to eight months, but Anthropic will also be offering new models by then that will push the frontier further, making it uncertain if others will be on par with the company.
In the long term, a persistent ban on Anthropic's tools would reduce the military's options for AI partners and increase reliance and lock-in risks for Grok, GPT and Gemini. The administration's heavy-handed tactics may even deter some AI startups with similar ideological views from working with the Pentagon, even in a narrow fashion.
Over the long term, the underlying dispute — the use of AI in military applications in the absence of clear rules — is only likely to increase in frequency as a flashpoint between the U.S. military and the technology sector as AI tools become more proficient. The pace of AI innovation, from agentic AI to physical AI, does not appear to be slowing down. The military applications that AI will unlock over the next three to five years will almost certainly be substantial, with fully autonomous weapons becoming cheaper and more practical. AI's ability to integrate information for surveillance will also grow significantly. Given the low likelihood of the federal government clarifying rules around such AI applications, popular concern, particularly among more progressive and left-wing companies and their employees, will grow.
Companies that choose to work with the Trump administration also risk inviting internal and external backlash. OpenAI’s new contract with the DoD, for example, could see key staff resign over what critics claim is the company’s failure to take a stand on AI safety. There have been calls to boycott OpenAI as well, though this is more likely to affect individual subscribers than the company’s lucrative business with organizations. Concerns over this backlash are likely, in part, what drove Altman to recently announce changes to the DoD contract.
For the United States more broadly, the administration's supply-chain designation and federal government use-phase-out remaining in place indefinitely — and especially if the government were to pressure contractors to reduce business ties to Anthropic — would undercut AI innovation. Anthropic itself could be significantly threatened by the decision, to the point where its ability to remain a developer of frontier AI models — or at least its ability to expand and commercialize them on a large scale — could become compromised, reducing overall competition among U.S. AI developers.
The heavy-handed tactics could also lead AI startups and companies to shift some of their development and resources overseas, particularly to AI-friendly jurisdictions such as Canada and the United Kingdom, though a company like Anthropic may find an incentive to expand its focus on the EU market, where its ideological views are more in line with political leaders'. The administration's tactics to blacklist an AI company could also undermine investment into the U.S. market over fears about increased politicization of the technology sector over the next three years, particularly if AI risks become more politically salient.
Finally, the United States' moves against Claude and the negative impacts they will have on the AI sector, both for military and civilian use, risk benefiting China in the AI race. Chinese AI labs are already closing the gap with their American counterparts. In some areas, like innovation on the structure of AI models and training algorithms, they equal or surpass their American counterparts. Their pursuit, however, has been largely hobbled by their inability to use some of the more advanced AI chips that American companies have unfettered access to.
By placing large restrictions on one of its leading AI companies and sending a chilling message to AI startups and investors across the industry, the United States risks slowing down its own advancement and making it easier for China to catch up and potentially pass the United States. While the administration's drive to use AI in military applications is, in part, driven by the goal of maintaining military technology dominance against China and other adversaries, absent a total war between China and the United States, the greater threat to China's AI advancement is the commercialization of AI, which the United States is seeking to halt Anthropic from being able to do.
