Business
Gist from The Wall Street Journal

Anthropic vs. the Pentagon: The Emails That Blew Up a Landmark AI Deal

Summarized July 2, 2026
Jump to key takeaways

Newly released court documents expose the detailed, months-long breakdown between Anthropic CEO Dario Amodei and Emil Michael, the Pentagon's undersecretary of defense for research and engineering, over how the military could use Anthropic's Claude AI models. The core dispute: Anthropic wanted explicit prohibitions on using its technology for fully autonomous weapons and domestic surveillance, while the Defense Department insisted it must retain flexibility to deploy AI across all lawful use cases — and refused to accept compliance language it deemed unworkable.

The email chain stretches from January through late February. Michael initially reached out to Amodei in mid-January following weeks of holiday delays, with negotiations that had started in late 2024. Amodei pushed back on surveillance applications, specifically flagging legal frameworks — such as foreign-intelligence statutes — that technically allow collection of Americans' communications. Michael countered that the Pentagon already follows the law, and that Anthropic was demanding compliance processes beyond what the department could realistically implement.

By February 26, Amodei formally rejected the Defense Department's latest proposal, calling it riddled with loopholes, and acknowledged the two sides had no path forward. The fallout was swift and dramatic: the very next day, Defense Secretary Pete Hegseth announced on social media that Anthropic would be designated a supply-chain risk, effectively barring Pentagon contractors from using Claude in their defense work. President Trump simultaneously directed federal agencies to stop using Anthropic's tools entirely.

A federal judge has since blocked enforcement of parts of those actions, but the government is appealing. Meanwhile, the damage is tangible: Michael has stated that roughly two-thirds of Pentagon operations that relied on Anthropic's models have already migrated to competing AI tools. Federal agencies scrambled immediately — Commerce Department officials were emailing colleagues about removing Claude from systems as late as 10 p.m. on March 3. Anthropic has sued over the supply-chain designation, framing the legal fight as necessary to protect its customers and partners while insisting it still wants to work productively with the government.

Key Takeaways

  • Amodei vs. Michael email chain drove Pentagon-Anthropic rupture
  • Core fight: autonomous weapons and domestic surveillance restrictions
  • Pentagon refused compliance language deemed 'unworkable' by DoD
  • Hegseth declared Anthropic a supply-chain risk the day talks collapsed
  • Two-thirds of Pentagon's Claude-using operations have since switched tools
  • Federal judge blocked enforcement; government appeal ongoing
  • Commerce Department rushed to purge Claude from systems by March 3
Read original article at The Wall Street Journal

Summarize any article in seconds

Gist is a free AI reader for your browser, iPhone, and Android. Get concise summaries and key takeaways from any article or podcast.

Get Gist — Free
⚡ Instant summaries 💬 Chat with articles 🔒 Privacy-first