Emails reveal Anthropic's standoff with Pentagon over autonomous weapons and surveillance AI

Anthropic and Pentagon Clash Over AI Model Usage

Court documents released Tuesday reveal the contentious email exchange that fractured the relationship between Anthropic and the Department of Defense earlier this year 1

1

2

2

. The 346-page filing exposes a fundamental disagreement over how the Pentagon could deploy Claude models, with Dario Amodei drawing firm ethical guardrails around autonomous weapons and domestic surveillance 1

1

.

The Anthropic and Pentagon conflict escalated in January when Emil Michael, the undersecretary of defense for research and engineering, reached out after weeks of silence. Michael pressed Amodei to accept the Pentagon's position that it could utilize the technology for "all lawful uses"—language that creates considerable wiggle room for military use of AI models 1

1

. Amodei maintained his stance, insisting on guardrails that would bar the technology from fully autonomous weapons systems and domestic surveillance tools.

Pentagon Rejects Ethical Boundaries

Michael's response was direct: the proposed restrictions were "just not workable" 2

2

. He warned Anthropic had "one more chance to align on core principles that would lead to legal language" before the relationship ended 1

1

. The Pentagon official also dismissed any attempt to distinguish between defensive and offensive weapons, stating "there is no distinction in our world between weapons that are defensive or offensive" 1

1

.

Amodei pushed back on the "all lawful uses" standard, pointing out that US law does permit domestic surveillance—precisely what Anthropic wanted to prevent 1

1

. He noted that the Pentagon's proposed language appeared to "completely remove our redlines" 1

1

. The next day, Defense Secretary Pete Hegseth designated Anthropic as a supply-chain risk, effectively ending negotiations 1

1

.

Questions of Conflict and Corporate Responsibility

Emil Michael's role in the negotiations has drawn scrutiny due to financial disclosures showing he held stock in xAI, an Anthropic competitor, alongside other AI investments 1

1

2

2

. This raises questions about AI governance and whether corporate responsibility can coexist with government contracts when officials negotiating access have competing financial interests.

A federal judge granted Anthropic a preliminary injunction in late March, calling the supply-chain risk designation "classic illegal First Amendment retaliation" 2

2

. However, an appeals court reversed that decision in April, and the case continues 2

2

.

Broader Implications for AI Ethics

The dispute tests whether AI companies can set ethical limits on government customers without losing contracts—a question central to Europe's debates over military and surveillance AI 2

2

. The EU AI Act wrestles with similar boundaries, and the standoff feeds sovereignty debates as European buyers weigh how much control US labs retain once their models enter national-security work 2

2

.

Anthropic has since steadied itself, nearing a deal this week to restore access to a restricted model 2

2

. The emails show how close the relationship came to permanent rupture and establish a precedent for how AI labs might navigate government demands that conflict with stated principles.