Anthropic on the Blacklist: An Unprecedented Standoff with the Pentagon

Open Conflict Between the Pentagon and Anthropic

The Pentagon has recently made a radical decision by placing the startup Anthropic on a blacklist, illustrating a growing divide between the Trump administration and part of the tech industry. This unprecedented measure was taken after Peter Hegseth, the U.S. Secretary of Defense, issued an ultimatum to Anthropic. The company was required to grant full access to its technology to the Department of Defense, within legal limits, or face contract termination and a ban on its technology from the administration. Dario Amodei, the head of Anthropic, refused, leading to this rupture.

Donald Trump's Accusations

Following Anthropic's refusal, Donald Trump expressed his anger on his Truth Social network, accusing the company of being "woke" and part of the "radical left." The Department of Defense then classified Anthropic as a risk to its supply chain, thereby prohibiting its use by any branch of the government. This classification, previously reserved for foreign companies like Huawei, marks a first for an American company.

Promising Collaboration Compromised

Anthropic was at the heart of the U.S. government's AI adoption strategy. Last July, the startup secured a $200 million contract from the Pentagon to develop AI applications for national security. While the government also collaborates with other AI players like Google, OpenAI, and xAI, Anthropic was the most utilized, particularly due to its integration with Palantir, and the only one deployed on classified systems. Its technology had been mobilized during the operation to capture Nicolás Maduro in Venezuela.

Internal Tensions and Departures at Anthropic

Relations began to deteriorate in early January after Peter Hegseth published a memo announcing a strategy to transform the U.S. military into an "AI-driven combat force," advocating for the use of this technology free from any "usage policy constraints." This approach, combined with the use of Anthropic's technology during the operation in Venezuela, caused internal unrest. On February 15, a researcher specializing in AI security left the company, illustrating the growing discomfort.

Pressure from the Department of Defense

Meanwhile, the Department of Defense intensified pressure on Anthropic and other companies to allow military use of their systems for "any legal purposes," including the development of weapons, intelligence, and battlefield operations. Anthropic's reluctance led Peter Hegseth to impose a deadline on Dario Amodei, and in the absence of compromise, he opted for a radical break.

The Trump Administration's Approach

The Trump administration, which ran a decidedly techno-libertarian campaign in 2024, has managed to attract many executives from Silicon Valley, disillusioned by Joe Biden's perceived anti-tech policies. Since its inception, it has advocated for unbridled AI adoption, opposing any regulation and promoting rapid technology adoption across all branches of government, including Defense. However, this approach goes too far for some voices within tech, including Anthropic, through its leader Dario Amodei, who regularly warns about the risks of unregulated AI.

The Dangers of Unregulated AI

Anthropic is not alone in its concerns about the rapid adoption of AI by the Pentagon. Gary Marcus, an American AI expert, shares these worries. He believes that current large language models are too prone to hallucinations and errors to be used for military purposes. "Generative AI is absolutely not reliable enough to make life-or-death decisions on a large scale," he writes on his blog. He warns that deploying these technologies without sufficient precautions could lead to disaster.

Alarming Study from King's College London

A recent study from King's College London, conducted by researcher Kenneth Payne and his team, offers little reassurance. They pitted three large language models—GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash—against each other in war games simulating international conflicts. The AIs, tasked with organizing a country's defense, opted for the use of nuclear strikes in 95% of scenarios. Some experts also suspect that a strike targeting a school in Iran, which killed 150 children, may have been linked to an AI error.

Fears of Increased Surveillance

There are also concerns that the Trump administration could use AI to enhance large-scale surveillance in the United States, for example, to improve the efficiency of mass deportations or monitor political opponents.

Sam Altman and the Turnaround

Sam Altman, who has a tumultuous relationship with Dario Amodei, seized the opportunity to strike a deal with the Pentagon after initially supporting Anthropic. While he claims to have secured safeguards from the Department of Defense, many experts doubt the robustness of this agreement. OpenAI is not yet approved for classified uses, partly because its technologies are not available via Amazon's cloud, which the U.S. government uses. However, a partnership between OpenAI and Amazon, with a $50 billion investment from the e-commerce giant, could change the game.

Calls for Regulation

Several AI experts, such as Timothy B. Lee, are urging Congress to legislate to limit certain uses and prevent a race to the bottom driven by major companies in the sector to secure the most lucrative contracts, even at the expense of ethics. However, it is unlikely that such a law will be passed before the midterm elections, as the current Republican majority is determined to follow Donald Trump's accelerationist approach.

A Dangerous Game for Trump

By targeting Anthropic and ignoring the concerns expressed by many AI professionals, including Sam Altman, regarding the lack of oversight of the technology, Trump is taking a considerable risk. Silicon Valley entrepreneurs may indeed fear being the next to suffer the same fate as Anthropic and decide that it is in their best interest to turn back to the Democrats, rather than live under the constant threat of such a Damocles sword.