Get Full Government Meeting Transcripts, Videos, & Alerts Forever!
Get email alerts on the Ai Governance topic
No spam. Unsubscribe anytime.
San Francisco DPH outlines AI policy, governance and use cases; emphasizes equity and vendor transparency
Summary
DPH presented an AI policy and governance plan that prioritizes equity, transparency and layered review; the department will buy AI solutions, has created an AI subcommittee, and is piloting partnerships (including a private GPT with UCSF) while flagging California laws such as AB 3030 that require AI disclosures in health care.
The San Francisco Department of Public Health presented its approach to artificial intelligence to the Health Commission, emphasizing governance, equity and vendor transparency as central commitments.
Chief Information Officer Eric Raffin said the department will generally "buy, not build" AI and will tie AI investments to measurable DPH objectives and process-improvement work. Raffin summarized eight guiding principles in DPH's AI policy, singling out equity, beneficence/non-maleficence, and transparency as priorities. "Transparency and accountabilityare the heavy lifting part," he said, and described an AI intake and review process that will require vendors to explain data sources and algorithm behavior before deployment.
Raffin announced an AI subcommittee beneath DPH's information governance structure and said the subcommittee has already begun reviewing requests through the new governance process. He described ongoing work with UCSF on a private generative-AI tool (a "private GPT" called Versa in the presentation) that would limit visibility to authorized institutional users and would require data-sharing agreements and legal review before DPH data could be used as prompts or training inputs.
On regulation, Raffin cited federal guidance gaps and pointed to recent California bills. He identified AB 3030 (described in the meeting as the Healthcare Services Artificial Intelligence Act) as requiring disclosures when AI is used in health care, and noted a separate state law requiring insurers to justify AI use in coverage eligibility decisions (referred to in the presentation as a recently passed Senate bill). Raffin said the state-level changes and fast-moving legislation will affect DPH policy and contracting.
The presentation included examples of current or near-term use cases: generative-AI pilots, document summarization, automated patient-portal responses, ambient-note solutions that draft clinical documentation from recorded visits, predictive analytics (reducing readmissions, sepsis detection), and imaging AI as diagnostic support. Raffin and commissioners discussed the high costs of some generative features, the need to prioritize equity in dataset selection and the importance of proof-of-concept contracting to evaluate vendor claims.
Commissioners asked about language in the policy that differentiates requirements indicated with "must" versus recommendations labeled "should." Raffin said the committee will consider stronger, more declarative language for core equity commitments while preserving flexibility for a wide range of pilot use cases. The commission asked for future updates focused on equity metrics and governance outcomes.
