HHS unveils ambitious AI strategy: what healthcare leaders need to know now
The U.S. Department of Health and Human Services released a comprehensive plan to expand its use of artificial intelligence. It's framed as a first step to streamline operations and create consistency across agencies. Between the lines, the document points to bigger goals: using AI for patient data analysis and accelerating drug development.
The plan adopts a "try-first" approach to raise productivity and give staff hands-on access to AI tools, including department-wide access to ChatGPT earlier this year. HHS reports 271 active or planned AI efforts in FY2024, with a projected 70% increase in 2025. Momentum is clear, but so are the privacy and safety questions.
What's in the plan
- Governance: Establish a framework to evaluate and manage risk.
- Shared tools: Build a suite of AI services available across the department.
- Workforce enablement: Train employees to use AI responsibly and effectively.
- R&D standards: Fund standards for AI in research and development.
- Public health and care: Integrate AI into public health programs and patient care.
The document highlights ambitions like delivering personalized, context-aware health advice by securely interpreting medical histories in real time. Some stakeholders are uneasy about potential data-sharing with large tech vendors, especially given past controversies over sensitive data handling, including the sharing of certain Medicaid data with immigration authorities.
The opportunity and the risk
Experts welcome progress while warning against speed without safeguards. Oren Etzioni praised the intent but cautioned that centralized data and fast rollout raise stakes when the data is health records. He pointed to the need for strong science, risk assessments, and transparency throughout development.
Darrell West at Brookings noted the plan promises better risk management but leaves implementation details thin. Key issues remain open: how sensitive data will be handled, what data sharing will look like, and how protections will extend to aggregated datasets used by AI tools.
Protections for individual records are clearer than for aggregated or derived data. The core challenge is balancing data use for better operations and outcomes with safeguards that keep patients and clinicians protected.
Practical actions for providers, payers, and public health teams
- Stand up AI governance now: Cross-functional board (clinical, compliance, security, legal, data) with clear decision rights.
- Set PHI rules: Default to data minimization, zero data retention by vendors, and field-level access controls.
- De-identification policies: Use Safe Harbor or Expert Determination and document methods. Reference HIPAA guidance here.
- Model evaluation: Validate for accuracy, bias, drift, and clinical safety. Require test sets that reflect your population.
- Human oversight: Keep a clinician or qualified reviewer in the loop for decisions that affect care.
- Procurement guardrails: Mandate BAAs/DPAs, data flow maps, SOC2/ISO evidence, red-teaming, and clear liability terms.
- Security and auditability: Log prompts, outputs, model versions, and access. Retain evidence for audits.
- Patient trust: Plain-language notices, consent where appropriate, and easy opt-outs.
- Workforce upskilling: Train clinicians and ops teams on safe AI use, prompt quality, and data hygiene. For structured learning by job role, see Complete AI Training.
- Measure outcomes: Track quality, safety events, equity impact, admin time saved, and cost-to-serve.
Near-term timeline to watch
- FY2024: 271 active or planned AI efforts inside HHS.
- 2025: Targeted 70% increase in projects; expect more pilots in patient engagement, claims integrity, and R&D tooling.
Also watch federal guidance shaping agency use of AI. For context on current federal direction, review the White House executive order on AI here.
Key questions to press vendors and agencies on
- What PHI leaves our environment? For how long? Who can see it?
- Is training on our data opt-in only? Can we disable data retention and model training?
- What's the model's provenance, update cadence, and deprecation plan?
- How are outputs validated, explained, and traced back to inputs?
- What is the human review and appeal process for high-stakes decisions?
- How are equity and accessibility measured and improved over time?
The bottom line
HHS is signaling scale, speed, and a push for practical AI in operations, research, and care delivery. The upside is real, but so are the privacy, safety, and accountability demands. Move fast on governance, training, and vendor standards so your teams can adopt AI with confidence-and patients remain protected.
Your membership also unlocks:
