Developing: Pentagon Gives Anthropic Friday Deadline to Drop AI Safety Guardrails — Or Face Blacklisting and Defense Production Act

According to sources familiar with the Tuesday meeting at the Pentagon, the conversation between Hegseth and Amodei was described as cordial in tone but unambiguous in substance. Amodei reportedly laid out Anthropic's longstanding position: the company supports the national security mission and is proud to have been the first AI firm deployed on classified networks, but it draws firm lines at involvement in autonomous kinetic operations — where AI makes final military targeting decisions without human intervention — and mass domestic surveillance.

Hegseth's response was blunt. According to CBS News, the Defense Secretary used an analogy: when the government purchases Boeing planes, the aerospace company has no say in how the Pentagon uses them. He argued the same principle should apply to Anthropic's Claude AI models. Sources told CNBC that if Anthropic fails to comply by 5 p.m. Friday, Hegseth threatened to label the company a "supply chain risk" — a designation typically reserved for foreign adversaries like Chinese telecom firms — which would require all Defense Department vendors and contractors to certify they do not use Anthropic's technology. Simultaneously, the Pentagon would consider invoking the Defense Production Act, a Korean War-era statute that allows the president to compel domestic companies to produce goods or services deemed critical to national security.

An Anthropic spokesperson offered a measured response: "We continued good-faith conversations about our usage policy to ensure Anthropic can continue to support the government's national security mission in line with what our models can reliably and responsibly do." The spokesperson added that Amodei "expressed appreciation for the Department's work and thanked the Secretary for his service" during the meeting.

The current standoff did not materialize overnight. Tensions began intensifying after reports emerged that Anthropic's Claude model was used — through a partnership with defense contractor Palantir — during the U.S. military operation that led to the capture of former Venezuelan President Nicolás Maduro in January. While the exact nature of Claude's role remains classified, Palantir's 2024 partnership announcement stated the AI could be used for "processing vast amounts of complex data rapidly" and "helping U.S. officials to make more informed decisions in time-sensitive situations."

According to NBC News, during a routine meeting between Anthropic and Palantir following the operation, a Palantir executive grew alarmed when an Anthropic employee appeared to question how the company's systems may have been used. Semafor reported that this exchange led to "a rupture in Anthropic's relationship with the Pentagon." A senior Pentagon official told NBC News that a senior Anthropic executive had contacted Palantir to inquire whether their software was used in the Maduro raid, and that the inquiry "implied that Anthropic might disapprove."

Anthropic has pushed back on the characterization that any single incident caused a breakdown. The company said it has not held out-of-the-ordinary discussions about Claude usage with partners, and that it found no violations of its policies in the wake of the Maduro operation. Nonetheless, the episode appears to have crystallized the Pentagon's view that Anthropic cannot be trusted to operate as a reliable military partner while maintaining independent ethical guardrails.

The Pentagon's pressure on Anthropic exists within a broader Trump administration campaign against what officials have termed "woke AI." White House AI czar David Sacks has publicly accused Anthropic of promoting ideologically biased technology because of its stance on regulation and safety. Hegseth and other administration officials have adopted the label as a catch-all critique of AI companies that maintain internal safety policies restricting military or government use.

AI researchers and policy experts have noted that the term "woke AI" lacks precise definition. NPR reported that experts describe it as "a nebulous and ill-defined term that Trump officials seem to use to describe any and all safety protections on powerful AI tools and the belief that AI chatbots have liberal bias baked into their models." The framing effectively conflates two distinct issues — political bias in chatbot outputs, and corporate policies limiting military applications — under a single politically charged umbrella.

Anthropic's competitors have largely acceded to the administration's demands. OpenAI, Google, and Elon Musk's xAI have all agreed to allow their AI tools to be used in any "lawful" scenarios. xAI was recently approved as the second company allowed to deploy models in classified settings, after agreeing to the Pentagon's terms. This dynamic has isolated Anthropic as the sole major holdout, making it both a symbolic target and a practical test case for the administration's authority over the private AI sector.

The Pentagon's threat to invoke the Defense Production Act represents an extraordinary escalation. The law, originally passed in 1950 during the Korean War, grants the president sweeping authority to direct private companies to prioritize government contracts and produce goods deemed essential to national defense. It has been invoked in recent memory during the COVID-19 pandemic to compel production of ventilators and personal protective equipment, but its use against a technology company over intellectual property and usage rights would be unprecedented.

A senior Pentagon official told NPR that such a move would compel Anthropic to allow its tools to be used by the military "if they want to or not." Legal experts say the application of the Defense Production Act to an AI company raises novel questions about whether the law's authority extends to compelling a software company to remove safety guardrails from its products, rather than simply requiring increased production of a physical good. The designation of Anthropic as a "supply chain risk" would carry its own severe consequences, effectively freezing the company out of not only direct government contracts but the vast ecosystem of defense contractors and subcontractors who rely on Pentagon business.

For Anthropic, the financial stakes are significant but not existential. The company recently closed a $30 billion funding round at a $380 billion valuation and reports more than 500 enterprise customers spending over $1 million annually. CEO Dario Amodei has pointed out that Anthropic's valuation and revenue have only grown since it began publicly pushing back against the administration. However, the company is also planning an initial public offering this year, and the uncertainty created by a confrontation with the federal government could complicate that process significantly.

Beyond the immediate contractual dispute, the Anthropic-Pentagon standoff raises a question that will define the coming era of military technology: should the companies that build advanced AI systems have any role in determining how those systems are used in combat?

Amodei has articulated his concerns in stark terms. In a January essay, he wrote: "My main fear is having too small a number of 'fingers on the button,' such that one or a handful of people could essentially operate a drone army without needing any other humans to cooperate to carry out their orders. I think we should approach fully autonomous weapons in particular with great caution, and not rush into their use without proper safeguards." Anthropic has also raised technical concerns, with sources telling CBS News that Claude is not immune from AI hallucinations and is not reliable enough to avoid potentially lethal mistakes — such as unintended escalation or mission failure — without human judgment in the loop.

The Pentagon's counterargument is straightforward: the military operates under the law, and lawful orders should not be subject to a private company's veto. Chief Pentagon spokesman Sean Parnell told NBC News that "our nation requires that our partners be willing to help our warfighters win in any fight. Ultimately, this is about our troops and the safety of the American people." Emelia Probasco, Senior Fellow at Georgetown University's Center for Security and Emerging Technology, urged resolution: "In my opinion, we should be giving the people we ask to serve every possible advantage. We owe it to them to figure this out."

Yet the precedent set here will resonate far beyond this single company. If the government can compel AI companies to remove safety measures using Cold War-era industrial statutes, it could fundamentally alter the incentive structure for AI safety research across the entire industry — potentially discouraging the very guardrails that many experts believe are essential as AI systems grow more powerful.