A Donated-Licence Pilot With 2,000 Practitioners on the Front Line
Two of the most-watched AI labs in the world have each handed over enterprise licences to a new pilot programme that will route frontier generative models directly to public health workers in ten U.S. jurisdictions. PULSE, short for Public Health Use Case and Learning Scaling Engine, is led by the Coalition for Health AI alongside Accenture, and is built to run from autumn 2026 through 2027, with the goal of producing reference playbooks the rest of the public health system can copy.
The licences cover capacity for up to 2,000 practitioners, and the Coalition for Health AI's leadership council will choose which state, local, tribal, and territorial agencies get seats. Accenture is taking the role of onboarding lead and will author the playbooks based on what the pilots surface, which gives the consultancy a quiet hand in setting the eventual national template.
Five Operational Tracks, No Clinical Decisions
The pilot deliberately stays on the operational side of public health. Instead of putting AI in front of patients, the five tracks target the paperwork, communications, and data plumbing that absorb most of a local health department's bandwidth.
The first track is built around biosurveillance and drug-wave prediction, with language models scanning overdose reports and open surveillance feeds for unusual patterns. The second applies AI to social-determinants mapping, extracting and geocoding health-related indicators from unstructured case notes and community surveys. The third tackles operations and community-feedback analysis: triaging resident messages, summarising meeting transcripts, and pulling themes from complaint data.
The fourth track builds a public communications capability wrapped around a multilingual translation hub, aimed straight at the equity gap that has dogged public health messaging since the COVID-19 response. The fifth wraps the most technical piece: a FHIR query engine for clinical-data retrieval that lets non-engineer staff ask natural-language questions of electronic health record systems.
The Quiet Governance Gap
What PULSE has not yet published is almost as striking as what it has. The programme has not laid out separate evaluation, privacy, security, or human-review frameworks for any of the five tracks. The Coalition for Health AI has not named the specific products, model versions, or configurations the two vendors will supply, and has not explained how OpenAI and Anthropic will be split across the pilots. The announcement does not say whether practitioners must approve generated public communications, verify translations, or validate retrieved clinical information before any of it is used in operations.
The CDC's own published considerations for generative AI in public health spell out expectations that PULSE has not yet committed to enforcing in measurable terms: human oversight on every output, scientific integrity, disclosure when AI appears in public-facing products, and security and privacy reviews for any non-public data. Whether the playbook phase closes that gap, or simply documents the pilots as they run, is one of the central questions practitioners will want answered before they sign on.
The Federal Backdrop
PULSE lands inside a federal environment that has been pushing hard on AI adoption in health. HHS published its strategic plan for AI across health, human services, and public health in January 2025, followed by an AI compliance plan in September 2025 that set a deadline of April 3, 2026 for high-impact AI use cases to meet minimum risk management practices. The department's own AI Use Case Inventory already lists live pilots, including automated metadata validation on data.cdc.gov running on Azure OpenAI services.
PULSE's distinctive contribution is its scale across jurisdictions, not its technology. For the first time, state and territorial health departments, county and municipal agencies, tribal authorities, Indian health organisations, and large-city health departments will all have a structured path to the same frontier commercial models, with shared onboarding and a single body of playbook material at the end. Whether that structure produces a coherent national practice, or a patchwork of vendor-flavoured case studies, depends on what CHAI chooses to publish when the pilots close.
What to Watch Through 2026 and 2027
Three signals will show whether PULSE becomes a genuine experiment or another vendor showcase. First, will CHAI publish track-specific technical, operational, privacy, and safety criteria before the autumn 2026 kickoff? Second, will the final playbooks be open public documents, or Accenture client material? Third, will participating jurisdictions retain the right to disclose how their staff actually used the tools, including the failures, after the pilots end?