Trump vs. Claude AI: The War Over Artificial Intelligence

Hours before the conflict with Iran, Trump and the U.S. also took a confrontational stance against Claude AI and its developer, Anthropic. The situation warrants a closer, detailed examination.

TABLE OF CONTENTS:

The Algorithmic Conflict

The United States is engaged in a different kind of war – one that is domestic, technological, bureaucratic, and ideological rather than military.

The administration of Donald Trump, alongside the Department of Defense, has effectively opened a front against one of the leading U.S. artificial intelligence firms, Anthropic, and its model Claude AI. Claude is now prohibited for federal use, and Anthropic has been labeled by the Pentagon as a “supply chain and national security risk.” The language is unusually strong, typically reserved for foreign contractors rather than domestic AI laboratories.

Amid heightened tensions with Iran, this internal confrontation takes on added gravity. The U.S. is signaling readiness for external conflict while simultaneously asserting control over algorithms, security standards, and the rules governing the defense AI market.

Ironically, until recently the U.S. government had a $200 million contract with Anthropic. The company presented itself as a lab focused on safety, advocating strict safeguards and ethical limits for its models – precisely the type of partner one would expect for integrating AI into military and analytical systems.

The situation has now reversed. The White House and the Pentagon have issued a complete ban on Claude AI within federal agencies, effectively cutting Anthropic off from the defense ecosystem. The president ordered the immediate cessation of all federal use of its solutions. The core issue is a disagreement over AI “guardrails”: the government and the developers differed on the limits of autonomy, acceptable use cases, and the level of oversight required.

As Anthropic faced isolation, another player emerged: OpenAI. The company quickly signed a new agreement with the Pentagon, effectively taking the space vacated by Anthropic. The rhetoric is familiar – safety, responsibility, oversight – but the emphasis has shifted. Now, flexibility in working with government requirements is as important as ethical principles.

This story is not merely about a change of contractor. It reflects a broader contest over who will shape the architecture of government AI in the U.S.: labs that prioritize safety over speed, or corporations willing to adapt to defense customer demands.

How did Claude evolve from a central tool in government research experiments to being considered a potential national security concern? Why did a multi-hundred-million-dollar contract fail to secure Anthropic’s position? And how did OpenAI manage to convince the Pentagon that its approach met the required standards?

This analysis examines the situation through five key points, avoiding sensationalism while acknowledging that the implications extend beyond a single contract.

Initial Agreement Between Anthropic and the U.S. Defense Department

For nearly two years, Anthropic was considered a key technology partner for the U.S. government. Its Claude model was integrated into several federal processes, and collaboration with the Pentagon appeared to be more than a temporary experiment, representing a structured partnership. At its peak, the company was valued at over $3.8 billion and secured a $200 million contract to adapt and deploy Claude for national security applications.

In practice, this was more than just another IT tool. The Pentagon positioned Claude as a foundational AI system authorized to operate within highly sensitive, classified cloud networks. The “Claude Gov” version became an internal standard, valued for its speed, intuitive interface, and ability to be rapidly configured for analytical tasks. U.S. military personnel increasingly relied on the system for planning, data processing, and auxiliary calculations.

Notably, Claude was also employed in operational scenarios. For example, it was used during campaigns exerting pressure on Nicolás Maduro’s regime in Venezuela. Its deployment in such contexts indicated a high level of trust, as the algorithm was permitted in processes where errors could have consequences beyond reputational damage, potentially affecting geopolitical outcomes.

This is where the core tension emerged. From the outset, Anthropic had built its reputation as one of the “most cautious” AI laboratories, emphasizing safety, ethical constraints, and safeguards against misuse. While these principles were seen as advantages in the civilian market, they became points of negotiation in the defense sector.

The tension centered on these negotiations. Military stakeholders sought broader system autonomy, more flexible application, and access to deeper model capabilities in specific scenarios. Anthropic, in contrast, insisted on maintaining restrictions and clearly defined usage boundaries. Formally, the discussions focused on technical parameters; substantively, they reflected a struggle over control – who determines the limits of what AI can do in military contexts: the government or the developer?

As a result, a partnership that initially appeared strategic gradually shifted into a conflict of interests. This divergence in perspectives on the role of AI in defense systems marked the starting point for the eventual split.

Safeguards Versus Military Flexibility

The partnership between the U.S. government and Anthropic did not collapse due to funding or technical failures. The critical issue was fundamental: who defines the boundaries for AI use in national security – the state or a private laboratory?

From the Pentagon’s perspective, the position was straightforward. Secretary of Defense Pete Hegset advocated for greater flexibility and expanded authority. Military leaders sought to remove some internal restrictions and obtain “unrestricted access” to Claude for any lawful defense-related task. The rationale was clear: if the government is responsible for security, it should determine both the tools and the rules governing their use.

However, Anthropic CEO Dario Amodei declined to revise the company’s core ethical framework. The company defined two clear “red lines” that it would not cross.

The first was a prohibition on using Claude for mass domestic surveillance of U.S. citizens.

The second was a restriction against integrating the model into fully autonomous weapon systems capable of making force deployment decisions without meaningful human oversight.

Anthropic emphasized that it supported the use of its model in lawful defense programs, but these two exceptions were non-negotiable. For the company, these limits reflected concerns about reputation, long-term accountability, and its philosophy of safe AI development.

The Pentagon, in turn, rejected the premise. The agency stressed that the boundaries for technology use are determined by U.S. law and decisions of elected authorities, not by the terms set by a private company. In other words, the government was not prepared to delegate a veto over strategic security tools to a developer.

As a result, a technical discussion evolved into a fundamental conflict over decision-making sovereignty. This disagreement marked a point of no return in the relationship between the U.S. defense establishment and one of the most “cautious” players in the AI market.

Restrictions and Supply Chain Risk

Once it became clear that neither side was willing to compromise, the conflict moved into the political arena. According to sources in Washington, the White House demanded that Anthropic fully accept the Pentagon’s conditions. The message was essentially clear: either remove the restrictions or exit the process.

The company did not comply and missed the deadline for reviewing the agreement. In response, the Trump administration took decisive and structured action, implementing two measures that changed Anthropic’s status within the federal ecosystem.

First, a federal restriction was imposed. The President ordered the immediate cessation of the company’s technology use across all federal agencies, with a six-month transition period during which agencies must fully replace its solutions with alternatives.

Second, the company was effectively blacklisted. Secretary of Defense Pete Hegset announced that Anthropic would be classified as a “supply chain risk” in the national security domain.

This designation in the U.S. had traditionally been applied primarily to foreign competitors, such as Huawei. In this case, however, it was applied to a domestic AI company. The implications could be substantive: following this decision, any defense contractor working with the military faces potential risks to its contracts if it continues collaboration with Anthropic.

In other words, Washington began treating a domestic AI developer almost like a potentially hazardous external supplier.

In response, Anthropic announced its intention to challenge the decision in court. The company argued that labeling it a “supply chain risk” was legally unfounded and set a dangerous precedent, potentially establishing a new model of government pressure on the technology sector.

Thus, a technical dispute over AI usage boundaries escalated into a broader conflict between the government and part of Silicon Valley, with legal, reputational, and strategic consequences for the industry as a whole.

Artificial Intelligence, Surveillance, and “Killer Robots”

Washington’s most extreme measures against Anthropic were driven by more than contractual disagreements. At the core were questions of military dominance and state sovereignty in the era of algorithms.

The Pentagon has publicly stated that the issue does not involve mass domestic surveillance or the deployment of “killer robots.” However, Secretary of Defense Pete Hegset emphasizes a different point: the Department of Defense must retain full flexibility on the battlefield, including the use of AI, and cannot allow corporate policies to constrain the government’s strategic decisions.

His argument is straightforward. In competition with authoritarian states, particularly China, excessive internal restrictions could create vulnerabilities.

Former President Donald Trump expressed a sharper view. In a post on the platform Truth Social, he accused Anthropic of attempting to impose its own rules on the military, prioritizing corporate conditions over the government’s mandate. According to him, the country’s direction should be determined by elected institutions, not “uncontrolled” technology companies.

At the same time, Trump emphasized that the government had no intention of breaking the law, but it would not allow frameworks established by a private company to override national security policy.

Anthropic and its critics view the situation differently. They warn that removing key safeguards could gradually blur the line between an auxiliary algorithm and an autonomous system capable of influencing force deployment decisions – a domain where human oversight is not merely desirable but essential.

Consequently, the conflict extends beyond a single contract. It reflects a clash between two approaches: the government, which demands maximum operational freedom amid global competition, and the developer, which seeks to establish constraints before the technology becomes irreversible. This divergence in perspectives on the future of algorithm-driven warfare ultimately served as the catalyst for the split.

Anthropic Out, OpenAI In

The market reacted almost immediately. Just hours after the restriction on Anthropic was announced, another player stepped in. OpenAI CEO Sam Altman announced a new agreement with the Pentagon to deploy its models within the Department of Defense’s classified networks. Effectively, OpenAI moved quickly to occupy the niche that had just been vacated.

Altman emphasized that the company is not abandoning safety principles but proposes a different mechanism for implementing them – specifically, through technical control tools directly agreed upon with the government. In other words, instead of rigid corporate “red lines,” there is now a contractual framework integrated into the agreement itself.

According to him, the deal includes two core commitments:

The models will not be used for domestic mass surveillance.
Human accountability in force deployment decisions will be explicitly maintained.

Altman stressed that these principles are not an OpenAI innovation; they are already reflected in U.S. law and are now formally embedded in the terms of cooperation with the Department of Defense.

In this way, OpenAI presented the Pentagon with a compromise: formally maintaining the same safeguards, but without a conflict over sovereignty. The government defines the boundaries through law and contract, while the company provides the technical implementation.

This flexibility appears to have been a decisive factor at a moment when the military AI market was being reallocated almost in real time.

Read also:

Trump vs. Claude AI: How the AI Debate Is Unfolding in the U.S.