Saltgrass Institute Responds to America's AI Action Plan

Last month, when the White House rolled out its “Winning the Race: America’s AI Action Plan,” I found myself asking, “Did I miss some pages?” I read it again—this time specifically searching for a vision around healthcare. Healthcare represents nearly 20% of the U.S. GDP, yet it received barely a mention in the nation’s AI playbook.

Over the years, I've worn many hats in the health tech space—CEO, board advisor, board chair, general punching bag. I’ve led tech reboots that moved two steps forward, one step back—for what felt like an eternity—often uncovering unexpected costs, delays, and complexities that no roadmap could fully predict.  I've held my breath through corporate recapitalization highs and lows, revised revenue strategies in the wee hours, and popped champagne after guiding companies successfully through IPOs. So, when I consider what must be going through the minds of other health tech leaders, investors, and entrepreneurs who are trying to determine how the Action Plan concretely influences their goals and visions, I imagine their responses fall somewhere on the scale between disappointment at the lack of attention to healthcare and relief not to be overly constrained.

As a representative of the Saltgrass Institute, a non-profit thought-leadership organization dedicated to responsible development and deployment of AI in healthcare, I want to unpack our observations of the Action Plan and outline some of the considerations health tech operatives will have to confront. Regardless of where they fall on the spectrum, the reality is that coordinated guidance will be required for AI to be well-integrated in such a regulated industry.

1. The Action Plan could more prominently feature healthcare delivery as a primary use case for AI

Going through the White House AI Action Plan, we found little guidance around AI in healthcare delivery — one of the most impactful, costly, and AI-ready sectors in the U.S. economy.

While the Action Plan mentions the need to pilot AI testbeds for verticals including healthcare delivery, it makes no mention of the Centers for Medicare & Medicaid Services (CMS), Office of the National Coordinator for Health Information Technology (ONC), or the Center for Medicare and Medicaid Innovation (CMMI) — the very agencies responsible for Medicare, Medicaid, and health IT standards.

Saltgrass sees the need for more guidance directed at CMS, ONC, and CMMI for how to coordinate infrastructure, regulation, and data-sharing frameworks that enable and encourage AI adoption at the point of care.

2. AI can supercharge value-based care if policy leaders would facilitate that

The Action Plan also misses an opportunity to address how AI can accelerate the shift to value-based care (VBC) — where providers are paid to improve health outcomes, not volume of services.

VBC has a well-documented track record of reducing costly administrative burdens in the healthcare setting, including identifying high-risk patients, supporting clinicians with documentation summarization, developing intuitive payment workflows, and supporting provider treatment decisions. The VBC model creates an ecosystem of efficiency that improves health access and health outcomes.

Leveraging AI to supercharge a shift toward broad VBC model use needs to be prioritized and road-mapped. This can be done if the White House incorporates guidance into the AI Action Plan. Specifically, the Action Plan should formalize federal pilot programs with accountable care organizations (ACOs), risk-bearing providers, and Medicaid/Medicare plans to test the most effective AI tools that achieve optimal cost and care.

3. Health AI needs real-world testing grounds

The plan mentions “regulatory sandboxes” but doesn’t address the specific needs of health AI, which must meet the standards of the US Food and Drug Administration (FDA), Health Insurance Portability and Accountability Act (HIPAA), and ONC.

For example, if a startup building an AI tool for early cancer detection needs access to real patient data, IRB-lite protocols, and a clear path to procurement in the hospital, that path today is murky at best. If the White House AI Action Plan can modernize HIPAA regulation and create clinical-grade validation environments for health AI that incorporates real-world data access, simplified ethics reviews, and procurement pathways into FDA/CDRH, ONC TEFCA frameworks, it will enable real-world validation of AI tools that can be more rapidly deployed.

4. Open-source model access ≠ clinical readiness

While promoting open-weight models is commendable, most health AI systems require domain-specific tuning, compliance, and performance thresholds to ensure accuracy and safety. Open-source generic parameters may not be appropriate. A good example is if the generic open-source model can summarize a patient chart, but it misses a medication allergy, the consequences could be life-threatening.

Saltgrass sees the need to build a federated evaluation ecosystem for healthcare-specific AI benchmarks, including for diagnostic accuracy, readmission prediction, medication safety and other indicators, which offer more tailored support to FDA, payers, and health systems.

5. Health data infrastructure needs more attention

The plan lacks thought and investment around EHR interoperability, synthetic data standards, and shared annotation tools, which are all essential for training and validating health AI.

In a practical context, an AI model trained on one hospital’s data may fail to work at another hospital due to inconsistent formats.

For these AI tools and infrastructure to be clinically useful, the Action Plan needs to specifically call for de-identified, longitudinal health data repositories, deeper AI model training to ensure the AI understands empathy, nuance, and critical thinking that happens in patient, clinician, and caregiver communication patterns, and also call for post-market surveillance to ensure the infrastructure quality and operations are optimal. 

6. Spelling out implementation risks and buyer barriers in health

Adoption barriers like cybersecurity, procurement workflows, IT integration timelines, and clinical governance are real in health systems, yet are barely acknowledged in the plan.

Saltgrass recommends that the Action Plan include commercial-readiness programs that mirror DoD’s Tech Readiness Levels (TRLs) for AI, tailored to healthcare delivery organizations.

7. Workforce development must specify clinicians and health care professionals

While electricians and data center workers are essential, the Action Plan needs to be more comprehensive of other sectors, specifically health. As is, the Action Plan leaves out frontline clinicians and healthcare operations teams, who are critical AI adopters, trainers, and gatekeepers.

We see an opportunity for federal programs to train and certify “AI-capable clinicians” who can safely deploy, audit, and improve AI tools in practice across the health landscape.

As the White House evolves alongside the field, the Saltgrass Institute will continue unpacking the thinking behind these policies, so innovators can use key takeaways to improve their creations and share their voices as industry players. Healthcare can’t afford to be an afterthought in the AI revolution.