Anthropic and Pentagon Restart Claude Talks After Military-Use Clash

Anthropic CEO Dario Amodei has reopened negotiations with the U.S. Department of Defense after earlier talks collapsed over how the military could use the company’s Claude AI models, according to CNBC.

The renewed discussions reportedly involve Emil Michael, the Pentagon’s under-secretary for research and engineering, after a breakdown late last week centered on restrictions Anthropic wanted to place on military use.

Why the Anthropic and Defense Department talks collapsed

The negotiations fell apart after disagreements over guardrails for Claude on defense networks.

Anthropic reportedly pushed for guarantees that the models would not support domestic surveillance or autonomous weapons systems. The Defense Department reportedly argued it needs the ability to use the technology for any lawful purpose.

Amodei reportedly said the talks unraveled after the Pentagon requested the removal of language referencing “analysis of bulk acquired data,” which Anthropic feared could enable large-scale surveillance.

Fallout inside Washington

After the breakdown, President Donald Trump directed federal agencies to stop using Anthropic tools.

Defense Secretary Pete Hegseth also threatened to designate Anthropic as a national security supply-chain risk, according to the report.

The Pentagon had previously awarded Anthropic a $200 million contract that would let Claude run on classified government networks, making it the first major AI model deployed in that setting, CNBC said.

OpenAI deal adds fuel to the dispute

The standoff intensified as OpenAI signed a separate Pentagon deal, which sparked backlash online.

OpenAI CEO Sam Altman later said the agreement may have moved too quickly and argued that officials should not label Anthropic a security risk, according to the report.

The dispute underscores a widening fault line in AI: companies want enforceable limits on how customers use powerful models, while governments want flexibility for security missions.

As talks restart, both sides now face a sharper question than contract size or deployment speed: who decides the boundaries when advanced AI moves into military and intelligence workflows?