Warren Opens Investigation Into Pentagon AI Deals

The senator calls Anthropic's blacklisting "retaliation" and demands answers about what surveillance protections OpenAI actually negotiated.

United States Capitol building dome against blue sky

TL;DR:

  • Senator Elizabeth Warren sent letters to Defense Secretary Pete Hegseth and OpenAI CEO Sam Altman on March 23, demanding answers about Anthropic's "supply chain risk" designation and OpenAI's Pentagon contract
  • Warren calls the Anthropic blacklist "retaliation" for the company refusing to remove safeguards against mass surveillance and autonomous weapons
  • The core allegation: The Pentagon "weaponized longstanding statutes intended to protect against genuine national security threats" to punish an AI company for its ethics policies
  • Warren is also probing OpenAI's deal, questioning whether it "lacks sufficient safeguards" and may enable the very surveillance Anthropic refused
  • Deadline for responses: April 6, 2026

Why Warren Calls It Retaliation

The March 23 letters lay out Warren's case: the Pentagon could have simply terminated its contract with Anthropic. Instead, it slapped the company with a "supply chain risk" designation, a label previously reserved for foreign adversaries like Huawei and ZTE.

"I am particularly concerned that the DoD is trying to strong-arm American companies into providing the Department with the tools to spy on American citizens and deploy fully autonomous weapons without adequate safeguards," Warren wrote to Secretary Hegseth.

The senator's language is blunt: the blacklisting "appears to be retaliation."

Anthropic's offense? Telling the Pentagon it wouldn't remove two contractual guardrails. The first prevented mass domestic surveillance of Americans. The second required human oversight for autonomous weapons systems.

The Pentagon's response was the supply chain designation, which carries severe consequences. Anthropic can't work with any federal agency. Companies that embed Claude in their products face pressure to rip it out. The reputational damage is immediate.

What Did OpenAI Actually Agree To?

Warren's investigation isn't just about Anthropic. She's demanding answers about what OpenAI negotiated when it signed its Pentagon deal: the one announced the same day Anthropic got blacklisted.

The senator's concern: OpenAI's contract "does not appear to include the same safeguards" that Anthropic insisted on keeping.

This echoes the EFF's March analysis, which called OpenAI's contract language "weasel words." The EFF flagged terms like "intentionally" and "consistent with applicable laws" as loopholes the government has exploited before. You can't "intentionally" surveil Americans if you're just "incidentally" collecting their data alongside foreign targets. And "consistent with applicable laws" is meaningless when current law allows buying surveillance data from brokers.

Warren wants to know: Did OpenAI actually negotiate meaningful protections? Or did it just sign a deal that lets the Pentagon do whatever's currently legal?

What Warren Is Demanding

The letters request specific answers by April 6, 2026:

From Secretary Hegseth:

  • Explain the Pentagon's rationale for using a supply chain risk designation against an American company over a contract dispute
  • Detail what specific safeguards the Pentagon asked Anthropic to remove
  • Clarify why the same concerns don't apply to OpenAI's contract

From Sam Altman:

  • Provide the specific language in OpenAI's Pentagon contract regarding surveillance safeguards
  • Explain whether OpenAI's deal includes the same restrictions Anthropic refused to drop
  • Detail any safeguards against domestic surveillance and autonomous weapons deployment

Warren isn't just asking nicely. This is a formal congressional investigation with document requests. The answers (or lack thereof) could set up subpoenas and hearings.

This Comes As Anthropic's Court Case Heats Up

Warren's investigation landed one day before Anthropic's preliminary injunction hearing in San Francisco federal court.

At that March 24 hearing, Judge Rita Lin pressed Pentagon lawyers hard. She called the government's actions "troubling" and questioned whether applying the supply chain statute to a domestic policy dispute was appropriate.

A ruling could come any day. If Judge Lin issues an injunction, Anthropic gets relief while the constitutional case proceeds. If she doesn't, the company remains blacklisted, and the message to other AI companies is clear: refuse the government's demands, and you're a national security threat.

Warren's investigation adds congressional pressure to the judicial proceedings. The Pentagon is now defending its actions on two fronts.

The Bigger Question

Warren's investigation highlights what the EFF has been saying for weeks: Americans' privacy shouldn't depend on which AI company happens to take an ethics stand.

Anthropic drew a line. OpenAI negotiated something else. Neither decision went through Congress, courts, or voters. The rules for AI surveillance are being written in contract negotiations behind closed doors.

If Hegseth and Altman respond with meaningful answers, we'll learn what the Pentagon actually asked for, and what protections actually exist. If they stonewall, Warren has the subpoena power to escalate.

Either way, April 6 is the next date to watch.

Sources

Published: March 25, 2026