GW Law Faculty Publications & Other Works

Document Type

Article

Publication Date

2026

Status

Accepted

Abstract

The Anthropic-Pentagon dispute has generated widespread commentary but fundamental confusion about whether contractors can restrict the government's use of their products. This article argues the question is not novel. The scope of permissible restrictions depends on the acquisition pathway, the contract type, and the negotiated terms. The article surveys the principal pathways through which the federal government acquires AI and examines OpenAI's published Pentagon contract language, which adopts an "any lawful use" standard conditioned on existing legal authorities. A critical tension emerges: although the contract facially permits broad use, OpenAI's retained architectural control over its cloud-only deployment and safety infrastructure may impose practical constraints exceeding those Anthropic sought through express contractual restrictions. The article concludes that the public debate has focused on the wrong question. The more consequential governance failure is the government's inability to secure adequate transparency, audit rights, and safeguards when procuring AI through commercial pathways not designed for technologies this complex and consequential.

GW Paper Series

2026-24

Download

Included in

Law Commons

Share

COinS