A landmark legal and ethical clash between the US Department of Defense and the artificial intelligence company Anthropic has reached a critical juncture, setting a precedent for how advanced AI will be governed in military hands and testing the limits of governmental power over Silicon Valley.
The Pentagon took the unprecedented step on 5 March 2026 of formally designating the American AI safety firm as a “supply chain risk,” a category typically reserved for foreign adversaries. The move, announced by Defense Secretary Pete Hegseth and formally communicated to Anthropic the previous day, marks a dramatic escalation in a long-running dispute and threatens to severely curtail the company’s ability to secure military contracts.
The core of the conflict is Anthropic’s refusal to allow its flagship Claude AI model to be used for two specific purposes: mass domestic surveillance and fully autonomous weapons systems. Company CEO Dario Amodei has stated these are non-negotiable “red lines” held since Anthropic’s founding, principles he argues are fundamental to its identity as a safety-conscious developer. In response, the Pentagon asserts it needs the ability to use Claude for “all lawful purposes” and will not allow a vendor to restrict the lawful use of what it deems a critical national security capability.
A Collision of Cultures and Timelines
According to Sarah Kreps, a professor at Cornell University’s Tech Policy Institute and former US Air Force officer, the feud exposes a fundamental cultural and procedural rift. The military, she notes, often cannot wait for a bespoke “military-grade” version of a rapidly evolving technology like AI and seeks to leverage commercially developed tools for their immediate utility. “The challenge for the military is that these technologies are so useful they can’t wait,” Kreps observed. “They need to act quickly because of how valuable these tools are, but it’s not surprising that they ran into cultural differences.”
This tension around “dual-use technology” is compounded by Anthropic’s deliberate branding. The company, co-founded by former OpenAI and Google Brain researcher Dario Amodei, has built its reputation on warning about AI’s existential risks and prioritising safety. This made its initial decision to engage with the Pentagon and, notably, with data analytics firm Palantir—a company deeply embedded in military and intelligence operations—seem paradoxical to some observers. Palantir, which integrates Claude into its platforms for uses like analysing operational data, has publicly stated that providing AI to the military is a moral obligation aimed at preventing war.
The “Black Box” Problem and a Legal Challenge
Anthropic’s specific fears hinge on what happens after its software is delivered. Unlike hardware, AI models can be repurposed. Once Claude is integrated into classified military systems, Anthropic would have no visibility into—or control over—how it is used. “It goes into not just a black box, but Black Ops and classified systems that are closed off,” Kreps explained. This creates a profound accountability gap, raising the spectre of the technology being used in ways that violate the company’s core principles under the blanket justification of national security.
The Pentagon counters that its own internal policies and existing US law already prohibit the very applications Anthropic fears. However, critics argue the department’s uncompromising demand for unrestricted usage could erode those very protections. Anthropic is now pushing back through the courts. Dario Amodei has called the supply chain risk designation “legally unsound,” arguing the relevant statute, 10 U.S.C. § 3252, is intended for foreign adversaries. Some legal experts have echoed this, describing the Pentagon’s move as a potentially dubious overreach of authority.
The dispute echoes past confrontations between technology giants and the state, most notably the 2016 standoff between Apple and the FBI over unlocking a terrorist’s iPhone. A key difference, as Kreps points out, is that “once you hand this [software] over to the military, you no longer need Anthropic’s approval to use it as you see fit.”
AI on the Battlefield: From Analysis to Autonomy
While the legal battle unfolds, the US military has already reportedly used Claude in operations, including strikes in Iran, highlighting the surreal reality of the Pentagon simultaneously designating a supplier a risk while deploying its product. This underscores the pervasive and growing role of AI in modern warfare, far beyond theoretical “red lines.”
Kreps, drawing on her intelligence background, notes AI’s primary value in defence lies in processing vast amounts of information to improve the “signal-to-noise ratio.” It excels at pattern recognition—identifying a specific type of naval vessel in satellite imagery, for instance. Platforms like Palantir’s are built for this very purpose, integrating disparate data streams to aid military decision-making.
The ethical terrain becomes far more precarious, however, in counter-terrorism or targeting scenarios where AI might be used to identify individual human targets with fewer concrete identifiers. “He could be a combatant, he could be a civilian,” Kreps said. “It’s not a naval vessel… where it’s harder to get that wrong.” This concern feeds directly into the debate over autonomous weapons and the assurance of a meaningful “human in the loop”—a key element of Anthropic’s refusal.
The outcome of this clash is being watched closely across the tech and defence sectors. It has already influenced competitors; OpenAI has reportedly struck its own deal with the military, though CEO Sam Altman has characterised it as “confusing,” and its agreement is said to include prohibitions on autonomous weapons and mass surveillance. The Anthropic case acts as a live test for a critical question: to what extent can—or should—private companies that develop foundational technologies dictate the terms of their use by the state in matters of national security? The answer will shape the future of AI in conflict and redefine the relationship between Silicon Valley and the Pentagon for years to come.
