White House officials reportedly frustrated by Anthropic’s law enforcement AI limits (arstechnica.com)

Semafor reports growing tension between the White House and Anthropic after federal contractors for agencies like the FBI and Secret Service encountered roadblocks using Anthropic’s Claude models for surveillance tasks. Anthropic’s usage policy explicitly prohibits domestic surveillance, and two senior White House officials told Semafor they fear the company enforces those rules selectively and uses vague language that can be broadly interpreted. The restrictions matter because in some classified workflows Claude is reportedly the only model cleared for top-secret work via AWS GovCloud; Anthropic also offers a specialized national security service and has a nominal $1 deal to provide services to federal agencies, while maintaining bans on uses like weapons development. The dispute highlights how vendor-level policy choices and cloud security clearances can shape operational access to advanced models. For the AI/ML community, it underscores a tension between corporate safety/ethics controls and government demand for flexible tooling, especially when a model is technically capable of analyzing classified material but contractually forbidden from certain domestic applications. The situation also comes amid competing procurement moves—OpenAI recently secured a large ChatGPT Enterprise pilot for federal workers—illustrating how policy settings, platform clearances (e.g., GovCloud), and vendor agreements will increasingly determine which models agencies can deploy in sensitive contexts.