HHS seeks AI pilot for 1,000 power users

For HHS, the AI question is moving from “can staff use a chatbot” to “what controls have to exist before advanced models touch mission workflows at scale.” According to NextGov, the department’s June 8 request for information seeks industry feedback on a short-term, fixed-price pilot that would give up to 1,000 authorized, portable HHS “power users” access to advanced AI capabilities, with an option to expand access to 10,000 users depending on what a vendor offers.

The RFI is aimed above basic chat and summarization. HHS wants users to test premium reasoning, long-context and agentic-capable models, and to see which functions can be used immediately, which need configuration or integration, and which require more work on security, privacy, records, accessibility or authorization before enterprise scaling. That is the right ordering. The dangerous federal AI failure mode is not that an agency runs a pilot. It is that a pilot becomes a platform before anyone can explain the control boundary.

This also fits HHS’s broader OneHHS AI push. The department’s 2025 AI strategy called for governance, risk management, user-centered infrastructure and “risk-proportionate controls” across AI projects, according to NextGov/FCW’s earlier coverage of the strategy (https://www.nextgov.com/artificial-intelligence/2025/12/hhs-releases-ai-strategy-united-new-onehhs-approach/409983/). This RFI is the procurement-shaped version of that promise: give high-end users room to test the tools, then force the evidence into a shared operational framework rather than pretending policy language alone will do the job.

For practitioners, the Monday work is not selecting the flashiest model. It is mapping where the model will sit, what data it can reach, what records obligations attach, who can authorize use, and what has to be logged before HHS lets the experiment become infrastructure.