Popular AI assistants‘ safety features—the ones preventing them from helping with weapons or surveillance—just became a federal liability. The Pentagon slapped Anthropic, maker of the popular Claude AI, with a rare “supply-chain risk” designation after the company refused to strip away guardrails that block autonomous weapons and domestic surveillance applications.
The March 5th blacklisting followed failed negotiations where Defense Secretary Pete Hegseth demanded “any lawful use” access to Anthropic’s AI models. When CEO Dario Amodei balked at removing restrictions, the Pentagon pulled the trigger on a designation typically reserved for foreign threats or companies with actual security vulnerabilities.
Legal Experts Call Pentagon Move “Dubious”
Defense officials admit no technical risks exist, raising questions about true motives.
Anthropic fired back with a federal lawsuit on March 9th, claiming the designation violates free speech and due process rights. Legal experts aren’t buying the Pentagon’s justification either. “This designation is ideologically driven, with no evidence of actual supply-chain risk,” a defense official told DefenseOne. The law requires proof of sabotage or backdoor vulnerabilities—neither exists here.
Market Split Creates Winners and Losers
OpenAI scores Pentagon contracts while Anthropic faces potential investor exodus.
The blacklisting immediately bars military contractors from using Anthropic’s services, threatening to cripple the company’s valuation by cutting off cloud access through AWS and Google. OpenAI, meanwhile, secured fresh Pentagon deals by accepting “human oversight” compromises that Anthropic rejected. It’s like watching Netflix and Disney+ compete, except the stakes involve lethal autonomous systems instead of streaming rights.
Your AI Tools Caught in Crossfire
Consumer AI safety features face unprecedented government pressure.
This fight extends beyond defense contracts to the AI tools millions use daily. The Pentagon’s “any lawful use” demand creates pressure for all AI companies to remove safety guardrails that prevent misuse. If successful, it could fragment the market into “ethical AI” for consumers and “compliant AI” for government—assuming those categories don’t eventually merge.
The precedent concerns AI researchers who’ve spent years building responsible safeguards into these systems. Your chatbot’s refusal to help with dangerous requests just became a liability rather than a feature.
