Anthropic Rejects Pentagon’s Final Offer, Escalates AI Standoff Hours Before Deadline

Anthropic rejected the Pentagon’s latest contract offer Thursday, bringing the AI company within 24 hours of a government deadline that could terminate its $200 million defense contract and potentially blacklist it as a supply chain risk. CEO Dario Amodei said the company ‘cannot in good conscience’ accept terms requiring unrestricted military use of its Claude AI model, according to CNBC.

The standoff centers on two specific restrictions Anthropic seeks to maintain: preventing Claude from conducting mass surveillance of Americans and barring its use in fully autonomous weapons systems that make targeting decisions without human oversight. Axios reported that contract language received overnight from the Pentagon ‘made virtually no progress’ on these safeguards, with compromise language paired with legal provisions that would allow restrictions to be disregarded.

The Escalation Timeline

The dispute intensified during a Tuesday meeting between Amodei and Defense Secretary Pete Hegseth at the Pentagon. According to NBC News, Hegseth presented a stark ultimatum: allow Claude to be used for ‘all lawful purposes’ by Friday at 5:01pm or face consequences including contract termination, supply chain risk designation, and potential invocation of the Defense Production Act.

The Pentagon sent what it termed its ‘last and final offer’ Wednesday night, according to a senior Pentagon official cited by CNBC. Pentagon spokesman Sean Parnell defended the department’s position, stating the Pentagon has no interest in illegal mass surveillance or autonomous weapons without human involvement, and is only requesting use for ‘all lawful purposes.’

Technical and Strategic Realities

Anthropic holds unique leverage: Claude is currently the only AI model deployed across Pentagon classified networks through a partnership with Palantir, making it critical for sensitive military operations. But that exclusivity is eroding. NBC News confirmed that xAI’s Grok signed a deal Monday agreeing to ‘any lawful use’ terms, while negotiations with OpenAI and Google for classified access are accelerating.

In December negotiations, Anthropic offered to allow Claude for missile and cyber defense applications, according to NBC News. That concession proved insufficient. Pentagon Undersecretary Emil Michael told CBS News the military had ‘made some very good concessions’ including written acknowledgment of federal laws restricting surveillance of Americans, characterizing Anthropic’s response as inadequate.

Amodei’s statement Thursday highlighted what he called contradictory Pentagon positions: ‘one labels us a security risk; the other labels Claude as essential to National Security.’ He maintained that frontier AI systems are ‘simply not reliable enough to power fully autonomous weapons’ and lack the judgment human troops exhibit, according to CBS News.

The Defense Production Act Wildcard

The Pentagon’s threat to invoke the 1950s-era Defense Production Act represents unprecedented territory for AI Regulation. The law typically compels manufacturing during national emergencies—it was used during COVID-19 for ventilators and masks. Applying it to software and intellectual property raises constitutional questions, according to legal experts cited by The Hill.

A ‘supply chain risk’ designation would be equally severe. Euronews noted such classifications are typically reserved for foreign adversaries, and would force defense contractors including Boeing and Lockheed Martin to assess their Anthropic exposure. The Pentagon reportedly contacted these contractors this week, according to Axios, laying groundwork for potential blacklisting.

Industry Implications

The standoff reveals a broader recalibration of AI company-government relations. OpenAI recently removed explicit ‘safety’ references from its mission language, while Google dropped pledges against using AI for weapons or surveillance last year, according to analysis in Opinio Juris. Nvidia CEO Jensen Huang called for a negotiated resolution but acknowledged ‘it’s also not the end of the world’ if talks fail, CNBC reported.

Pentagon rhetoric has been pointed. Michael called Amodei a ‘liar’ with a ‘God-complex’ on X Thursday evening, adding that the CEO ‘wants nothing more than to try to personally control the US Military,’ according to CBS News. The comment reflects deeper ideological friction: Defense Secretary Hegseth declared in January that the Pentagon will not employ AI models that ‘won’t allow you to fight wars,’ framing company safeguards as ‘woke’ constraints.

What to Watch

The Friday 5:01pm deadline will clarify whether the Pentagon follows through on threats or finds an eleventh-hour compromise. Amodei stated the company would ‘work to enable a smooth transition to another provider’ if offboarded, avoiding disruption to military operations. But several factors complicate a clean break: Claude’s integration into classified systems through Palantir, the lack of immediate replacements with equivalent capability and security clearances, and potential legal challenges to Defense Production Act invocation.

If the Pentagon proceeds with a supply chain risk designation, expect litigation. Former Justice Department liaison Katie Sweeten told CNN the contradictions—labeling a company both a security risk and essential enough to compel via DPA—could undermine the designation’s legitimacy, making it appear punitive rather than security-based.

The outcome will establish precedent for how much control AI companies retain over model deployment versus government authority to commandeer technology deemed critical to national security. For Anthropic, the $200 million contract represents less than 2% of its $14 billion revenue, according to NPR, but the reputational stakes extend far beyond: Can a company market itself on safety principles while maintaining defense partnerships, or will the industry bifurcate between commercial AI providers and defense-cleared vendors willing to accept unrestricted terms?