A nonprofit called the Coalition for Health AI launched a program last week to test AI tools from OpenAI and Anthropic inside US public health agencies. The program, called PULSE, will run across 10 state, local, tribal, or territorial agencies starting this autumn. Results are expected to be published in 2027 as practical guides that other agencies can follow.
The setup is straightforward. OpenAI and Anthropic have each donated enterprise licences, giving up to 2,000 public health workers access to their tools at no cost. Consulting firm Accenture is handling onboarding and will help turn what the pilots learn into written guidance. The work is organised around five areas: tracking disease outbreaks and predicting drug crises, mapping social factors that affect health, handling community feedback and operational efficiency, producing public communications and translations, and pulling clinical records automatically.
The timing is not accidental. US public health agencies are under severe financial pressure right now. The federal government clawed back over $11 billion in COVID-era grants from state and local health departments, and the proposed CDC budget for the coming year would cut the agency's funding roughly in half compared to 2024. Departments across the country have been laying off staff, reducing lab capacity, and cancelling vaccination programmes. When a vendor or nonprofit offers free tools promising efficiency gains, that offer lands differently when you have just sent 170 employees home.
That context is useful for reading the launch language carefully. Nearly 40 percent of local health departments were not yet using AI at all, according to data cited in the announcement. The programme is positioned as a way to help those agencies build practical experience before committing to anything larger. That is a reasonable goal. The concern is that agencies under financial stress may move faster toward relying on AI outputs than the pilots were designed to support.
The privacy and data questions are the part that deserves the most attention. The announcement does not specify what data the 10 pilots will actually use. It does not say whether patient records will be involved, whether those records will be identifiable or anonymised, or who must review AI outputs before any action is taken. For tasks like drug-wave prediction or automated clinical data retrieval, those are not minor details.
Both OpenAI and Anthropic offer enterprise configurations that do not use customer data to train their models, and Anthropic's API does offer HIPAA-ready arrangements for organisations handling protected health information. But those options only apply if they are explicitly set up and contracted. The PULSE announcement does not confirm that those arrangements are in place for any of the pilots, and it does not set out retention rules, access controls, or audit requirements for the five use cases.
There is also a broader political layer. The US government has taken a deregulatory stance on AI, and a separate executive action earlier this year placed restrictions on government use of Anthropic products in defence contexts. Public health agencies are not the Department of Defense, but any organisation watching this programme should note that the legal and political ground under AI procurement is shifting quickly.
None of this means the programme is a bad idea. Public health agencies need to understand what these tools can and cannot do in real operating conditions, and a structured pilot with published results is a more honest approach than quiet individual deployments with no shared learning. The 2027 playbooks will be worth reading closely, because that is when the real story will be visible: whether the pilots produced genuine operational improvements, what broke, and whether the guidance is specific enough to be useful to an agency that looks nothing like the 10 that participated.