Pentagon Signs Seven AI Deals to Break Dependence on Anthropic
The Pentagon signed AI agreements with seven major technology companies Friday, expanding its classified network access. Its legal standoff with Anthropic shows no signs of resolution.
The deals cover xAI, OpenAI, AlphabetGOOGL, AmazonAMZN, MicrosoftMSFT, NvidiaNVDA, and startup Reflection AI.
Each company agreed to let the Defense Department use their technology for "any lawful use," a standard Anthropic has declined to accept.
Anthropic was previously the only AI model available on Pentagon classified networks.
The Pentagon said the agreements accelerate the transformation toward establishing the US military as an AI-first fighting force.
The cloud deals with Amazon and Microsoft carry less operational weight than the model agreements with OpenAI and Google, per the New York Times.
Model providers determine how AI gets applied in actual combat operations, not cloud hosts.
Potential military uses include generating target lists during active conflicts and processing large volumes of intelligence data, per the Defense Department.
The Anthropic Dispute
Anthropic and the Pentagon are in federal court after the Defense Department labeled the company a supply chain risk. It was a novel application of government contracting authority.
President Trump ordered agencies to cut ties with Anthropic by mid-2026, but intelligence analysts still depend on its older models for day-to-day work.
The new deals were partly intended to pressure Anthropic into accepting the broad Pentagon standard.
At least some newly signed companies received contractual safeguards around autonomous drones and domestic surveillance, similar to what Anthropic had sought, per the same report.
The Pentagon says it doesn't intend to use AI for drone piloting or domestic surveillance but has declined to put that commitment in writing.
A further complication is Anthropic's newest model, Mythos, which the White House is currently blocking from wider commercial distribution.
Mythos is capable enough at finding software vulnerabilities that Anthropic has not released it publicly. Early access is limited to select security researchers in the US and UK.
The White House cites cybersecurity risks from broader access, even as officials frame the situation as balancing innovation with national security.
Mythos's capabilities make a full Anthropic cutoff increasingly unlikely despite the standing executive order, per the same report.
The Pentagon's stated goal across all new agreements is preventing vendor dependence while pushing the military toward a single unified AI standard.