Anthropic Reopens Pentagon Talks After Standoff Over AI Safeguards

Anthropic, which recently made headlines for declining certain terms in a proposed AI agreement with the US Department of War (formerly Department of Defence), is now back in negotiations after risking potential exclusion from the military’s supply chain.

CEO Dario Amodei had previously resisted aspects of a defence deal, citing concerns over AI reliability and its readiness for use in mass surveillance or developing fully autonomous weapons systems. However, the company is now reportedly in discussions with Emil Michael, the US under-secretary of defence for research and engineering, to establish clearer terms governing how the Pentagon may use its AI models, according to a Financial Times report.

If finalised, the agreement would allow the US military to continue using Anthropic’s AI tools and help the company avoid being formally designated a “supply chain risk”—a label Defence Secretary Pete Hegseth threatened, which would’ve effectively excluded it from competing for contracts or subcontracts.

Last week, Anthropic drew a firm line over the military application of its technology. The company refused to remove two contractual safeguards: a prohibition on using its models for mass domestic surveillance and restrictions on deploying fully autonomous weapons systems.

In a blog post, Amodei argued that AI systems are not sufficiently reliable to operate without meaningful human oversight. Despite this stance, Anthropic has now returned to the negotiating table, although no formal agreement has yet been confirmed.

While the dispute risked straining Anthropic’s standing with US defence authorities, its Claude chatbot has continued to gain popularity among users.

By contrast, OpenAI faced significant backlash after striking its own agreement with the Pentagon. A “Cancel ChatGPT” campaign spread across social media platforms, contributing to a reported 295% surge in uninstalls last week.