Who Controls AI?

What It Covers

Anthropic's public standoff with the Pentagon over Claude's use in autonomous weapons and domestic surveillance culminates in President Trump ordering all federal agencies to cease using Anthropic technology, while OpenAI simultaneously negotiates a separate DOD agreement preserving its own stated safety red lines around the same two issues.

Key Questions Answered

• •AI Vendor Leverage: Private AI companies holding critical infrastructure contracts face an asymmetric power dynamic — governments can designate vendors as "supply chain risks" under 10 USC 3252, triggering contractor-wide bans. Anthropic's case shows this designation requires a formal risk assessment, written national security determination, and congressional notification before taking legal effect, meaning immediate declarations carry limited enforceability.
• •Red Line Negotiability: OpenAI secured DOD agreement preserving two specific prohibitions — domestic mass surveillance and autonomous lethal weapons without human oversight — by building a custom safety stack rather than relying on vendor-side terms of service. This framework, where the government controls technical guardrails internally, may represent the replicable model for AI companies seeking military contracts without ceding ethical oversight entirely.
• •Supply Chain Designation Scope: Hegseth's declaration that no Pentagon contractor may conduct any commercial activity with Anthropic would, if enforced literally, require Amazon, Microsoft, and Google — all DOD contractors — to sever ties with Anthropic. Since Anthropic distributes models through AWS, Google Cloud, and Azure, full enforcement would functionally eliminate Anthropic's ability to serve its models at scale.
• •Brand Risk Asymmetry: Claude reached the number two App Store position during the controversy, suggesting consumer sentiment shifted toward Anthropic. OpenAI faces a distinct reputational risk: if a simplified narrative linking OpenAI to DOD surveillance hardens among politically concentrated liberal user bases on TikTok and Instagram, switching behavior could compound faster than factual corrections can counter it.
• •Geopolitical Control Framing: The core unresolved tension is structural — democratic societies cannot force private AI companies to comply, yet cannot allow those companies to constrain military operations, while authoritarian competitors integrate AI into military chains with zero civilian oversight. Any AI company building in the US should factor government contract terms, not just commercial terms, into their foundational governance decisions.