The Emails That Tell the Real Story

For months, the clash between Anthropic and the Pentagon was reported as a dispute over access — the government wanted Claude's full capabilities, Anthropic was holding back. Court documents released this week, first reported by the Wall Street Journal, tell a very different story.

The emails trace a tense exchange between two principals: Anthropic CEO Dario Amodei and Emil Michael, the Pentagon's Undersecretary of Defense for Research and Engineering. What they reveal is that the fight was never about access. It was about whether an AI company can set ethical limits on a government customer.

The Red Lines

Amodei's position was consistent throughout months of negotiations. Two use cases were non-negotiable: fully autonomous weapons and domestic mass surveillance. These are the same boundaries he described in a CBS News Sunday Morning interview — 'Case number two is fully autonomous weapons,' he said, defining them as 'weapons that fire without any human involvement.' On surveillance, he warned that AI could make large-scale analysis of bulk personal data newly feasible in areas where 'we need to have a conversation.'

The Pentagon wanted something far broader: models cleared for 'all lawful uses.' The problem, Amodei pointed out, is that US law does permit domestic surveillance. Accepting the Pentagon's formulation, he wrote, would 'completely remove our redlines.'

'Just Not Workable'

Michael's response was blunt. The guardrails were 'just not workable,' he wrote. He offered Anthropic 'one more chance to align on core principles' before the two sides split. Most revealing was his dismissal of Amodei's distinction between defensive and offensive weapons: 'There is no distinction in our world between weapons that are defensive or offensive.'

Michael asked for Amodei to join a final call; he was unavailable. Minutes after the deadline elapsed, Defense Secretary Pete Hegseth designated Anthropic a supply-chain risk — a label usually applied to firms tied to foreign adversaries. The move turned a contractual dispute into a constitutional crisis.

Retaliation, Not Security Policy

A federal judge agreed. In late March, she granted a preliminary injunction, calling the blacklisting 'classic illegal First Amendment retaliation.' An appeals court reversed the decision in April, and the lawsuit continues. But the emails make one thing clear: the Pentagon wanted Claude available for autonomous weapons and mass surveillance, Amodei refused both, and the blacklisting followed.

Why It Matters

The dispute is the defining case study of whether an AI company can maintain ethical boundaries in government contracts. The questions it raises — can a lab say no to a weapons customer? Can a regulator own shares in the companies it oversees? — will define AI governance for the rest of the decade. For now, Fable 5 is back online, the lawsuit is ongoing, and the underlying question of who decides how the US military uses frontier AI remains unresolved.

Sources