US Public Health Agencies Pilot OpenAI and Anthropic AI

Public health departments across the United States are embarking on a novel initiative to rigorously test generative artificial intelligence tools. This program, dubbed the Public Health Use Case and Learning Scaling Engine (PULSE), is a collaborative effort involving the Coalition for Health AI (CHAI), leading AI developers OpenAI and Anthropic, and global consulting firm Accenture. The initiative aims to evaluate the practical application and effectiveness of advanced AI in critical public health scenarios, paving the way for broader adoption and informed implementation guidance.

PULSE will facilitate trials across ten distinct state, local, tribal, or territorial jurisdictions. The overarching goal is to generate actionable implementation blueprints for public health agencies contemplating similar AI deployments. To kickstart this ambitious undertaking, OpenAI and Anthropic have generously donated ten enterprise licenses, each capable of supporting up to 2,000 public health practitioners. Accenture will play a pivotal role in managing participant onboarding and developing comprehensive “playbooks” based on the insights gleaned from these trials.

This program offers public health professionals direct access to enterprise-grade AI solutions from both OpenAI and Anthropic. While the specific products, model versions, or configurations to be utilized remain undisclosed by CHAI, the initiative promises a structured approach to understanding how these cutting-edge tools can be leveraged. Dr. David Lakey, a former Texas health commissioner, aptly summarized the core principle, stating, “Every major technological transformation succeeds or fails based on trust, governance, and execution. PULSE will support agencies in this endeavor, and is specifically designed for practical implementation.”

Five Public Health Use Cases Under Scrutiny

The leadership council of the Coalition for Health AI will be instrumental in selecting the participating jurisdictions and assigning practitioners to specialized teams focused on five key use cases. These areas of focus are designed to address some of the most pressing challenges in public health:

• Biosurveillance and Drug-Wave Prediction: Leveraging AI to monitor disease outbreaks and predict the spread of illegal drug usage, enabling proactive interventions.
• Social Determinants of Health (SDoH) and Mapping: Utilizing AI to identify and map the complex social and economic factors that impact community health outcomes, informing targeted public health strategies.
• Operations and Efficiency, and Community-Feedback Analysis: Employing AI to streamline public health operations, improve resource allocation, and analyze community feedback to enhance service delivery.
• Public Communications and a Multilingual Translation Hub: Developing AI-powered tools to disseminate critical public health information effectively and create a robust translation hub to reach diverse linguistic communities.
• Automated Clinical-Data Retrieval and a FHIR Query Engine: Exploring AI’s capacity to automate the retrieval of clinical data using the Health Level Seven (HL7) Fast Healthcare Interoperability Resources (FHIR) standard, thereby improving data accessibility and analytical capabilities.

While the National Institute of Standards and Technology’s (NIST) AI Risk Management Framework emphasizes evaluating AI systems based on their intended use, operating environment, affected parties, and potential consequences, CHAI has yet to publish separate, specific evaluation, privacy, security, or human-review requirements tailored to these five PULSE use cases. FHIR, a vital HL7 standard, facilitates the electronic exchange of healthcare information between compatible systems. The inclusion of an automated clinical-data retrieval and FHIR query engine within PULSE presents an intriguing opportunity, though the precise role of generative AI in this workflow—whether in query generation, record retrieval, summarization, or a combination thereof—remains to be detailed. Furthermore, the protocols for verifying the accuracy of AI-generated queries, completeness of retrievals, or validity of summaries before operational use are not yet defined, underscoring the need for robust human oversight mechanisms.

The nature of the data to be used in these pilots—whether identifiable, de-identified, synthetic, or aggregated—is also a critical consideration. While the Health Insurance Portability and Accountability Act (HIPAA) mandates the protection of electronic health information, its applicability will vary depending on the specific participating organization, the data involved, and the function being performed. Both OpenAI and Anthropic have affirmed that their enterprise services, by default, do not use inputs and outputs from customers for model training. However, the specific configuration of these PULSE deployments, including data retention periods, access controls, audit arrangements, storage requirements, and rules for submitting protected health information, are yet to be elucidated. Felipe Millon, OpenAI’s head of government go-to-market, emphasized the program’s intent: “We believe AI should be useful, safe, and accessible to the people tackling society’s most important challenges.” The donated licenses are intended to facilitate this structured evaluation process.

Governance and Evaluation Frameworks Are Key

The pilot programs are slated to commence in the autumn of 2026, with CHAI anticipating the release of the comprehensive playbooks in 2027, serving as a valuable reference for other public health agencies. A significant aspect of this initiative lies in its evaluation methodology. CHAI has not yet detailed the specific metrics that will be employed to assess the success of the pilots, nor has it clarified whether each use case will be evaluated under distinct technical, operational, privacy, and safety criteria. The process for reviewing AI-generated outputs, such as approving public communications, verifying translations, validating retrieved clinical information, or scrutinizing biosurveillance and drug-wave predictions before their operational deployment, also requires further definition.

NIST’s guidance strongly advocates for identifying AI functions that necessitate human oversight and for training users to comprehend system performance and limitations. Their generative AI guidance further encompasses essential areas like testing, validation, monitoring, documentation, privacy, and management oversight. Elizabeth Kelly, Anthropic’s head of beneficial deployments, highlighted the program’s commitment to responsible AI integration: “Public health teams are being asked to do more with less, and AI can help—as long as it’s brought in with care and the right guardrails.” She added that PULSE aims to empower practitioners to test these tools within their own environments, with privacy, governance, and responsible-use measures embedded from the outset.

Data from the National Association of County and City Health Officials suggests that a substantial portion of local health departments—nearly 40%—are not currently utilizing AI. CHAI notes that many of these departments express keen interest in revising workflows and enhancing operational efficiency. PULSE aims to address this by providing enterprise licenses, onboarding support, peer communities, and implementation playbooks. However, specific minimum requirements for staffing, infrastructure, interoperability, or cybersecurity for participating jurisdictions have not been specified. Eligible participants encompass a wide range of public health entities, including state and territorial health departments, county and municipal agencies, tribal authorities, Indian health organizations, and large city health departments.

The program’s ambition is to distill findings from these ten pilot jurisdictions into guidance for broader application. The playbooks will need to meticulously account for variations in agency size, existing technical systems, legal responsibilities, staffing levels, and procurement processes. Crucially, the announcement does not specify whether outputs from biosurveillance, drug-wave prediction, or clinical-data retrieval will be solely for testing purposes, presented to staff for review, or directly integrated into operational workflows—a decision that has significant implications for risk management and operational impact.

PULSE is an integral component of CHAI’s broader efforts to establish robust governance standards for AI in healthcare. In May, the organization revealed plans to develop guidance across eight governance domains through collaborative workshops and working groups involving over 150 healthcare AI stakeholders. CHAI has already begun publishing playbooks on organizational AI policies, governance structures, and internal resource management, with further guidance on other aspects of healthcare AI management anticipated. Separately, CHAI has collaborated with The Joint Commission on governance playbooks aligned with its voluntary Responsible Use of AI in Healthcare certification, though it remains unstated whether PULSE participants will be assessed under this certification.

Dr. Brian Anderson, CEO of the Coalition for Health AI, pointed out that many public health agencies entered the COVID-19 pandemic with limited technological investment, underscoring the program’s timeliness. He believes PULSE will equip these agencies with practical AI experience before widespread implementation. Dr. Ashish Jha, a former White House COVID-19 response coordinator, offered a forward-looking perspective: “We know AI is going to reshape how we deliver public health—the question is whether we do it thoughtfully or not.” The PULSE program is poised to identify effective AI applications and meticulously document these findings for the benefit of the entire public health community.