The White House released “America’s AI Action Plan” last week, which outlines various federal policy recommendations designed to advance the nation’s status as a leader in international AI diplomacy and security. The plan seeks to cement American AI dominance mainly through deregulation, the expansion of AI infrastructure and a “try-first” culture.

Here are some measures included in the plan:

• Deregulation: The plan aims to repeal state and local rules that hinder AI development — and federal funding may also be withheld from states with restrictive AI regulations.
• Innovation: The proposal seeks to establish government-run regulatory sandboxes, which are safe environments in which companies can test new technologies. 
• Infrastructure: The White House’s plan is calling for a rapid buildout of the country’s AI infrastructure and is offering companies tax incentives to do so. This also includes fast-tracking permits for data centers and expanding the power grid.
• Data: The plan seeks to create industry-specific data usage guidelines to accelerate AI deployment in critical sectors like healthcare, agriculture and energy.

Leaders in the healthcare AI space are cautiously optimistic about the action plan’s pro-innovation stance, and they’re grateful that it advocates for better AI infrastructure and data exchange standards. However, experts still have some concerns about the plan, such as its lack of focus on AI safety and patient consent, as well as the failure to mention key healthcare regulatory bodies.

Overall, experts believe the plan will end up being a net positive for the advancement of healthcare AI — but they do think it could use some edits.

Deregulation of data centers

Ahmed Elsayyad — CEO of Ostro, which sells AI-powered engagement technology to life sciences companies — views the plan as a generally beneficial move for AI startups. This is mainly due to the plan’s emphasis on deregulating infrastructure like data centers, energy grids and semiconductor capacity, he said.

Training and running AI models requires enormous amounts of computing power, which translates to high energy consumption, and some states are trying to address these increasing levels of consumption. 

Local governments and communities have considered regulating data center buildouts due to concerns about the strain on power grids and the environmental impact — but the White House’s AI action plan aims to eliminate these regulatory barriers, Elsayyad noted.

No details on AI safety

However, Elsayyad is concerned about the plan’s lack of attention to AI safety.

He expected the plan to have a greater emphasis on AI safety because it’s a major priority within the AI research community, with leading companies like OpenAI and Anthropic dedicating significant amounts of their computing resources to safety efforts. 

“OpenAI famously said that they’re going to allocate 20% of their computational resources for AI safety research,” Elsayyad stated.

He noted that AI safety is a “major talking point” in the digital health community. For instance, responsible AI use is a frequently discussed topic at industry events, and organizations focused on AI safety in healthcare — such as the Coalition for Health AI and Digital Medicine Society — have attracted thousands of members. 

Elsayyad said he was surprised that the new federal action plan doesn’t mention AI safety, and he believes incorporating language and funding around it would have made the plan more balanced.

He isn’t alone in noticing that AI safety is conspicuously absent from the White House plan — Adam Farren, CEO of EHR platform Canvas Medical, was also stunned by the lack of attention to AI safety.

“I think that there needs to be a push to require AI solution providers to provide transparent benchmarks and evaluations of the safety of what they are providing on the clinical front lines, and it feels like that was missing from what was released,” Farren declared.

He noted that AI is fundamentally probabilistic and needs continuous evaluation. He argued in favor of mandatory frameworks to assess AI’s safety and accuracy, especially in higher-stakes use cases like medication recommendations and diagnostics.

No mention of the ONC

The action plan also fails to mention the Office of the National Coordinator for Health Information Technology (ONC), despite naming “tons” of other agencies and regulatory bodies, Farren pointed out. 

This surprised him, given the ONC is the primary regulatory body responsible for all matters related to health IT and providers’ medical records. 

“[The ONC] is just not mentioned anywhere. That seems like a miss to me because one of the fastest-growing applications of AI right now in healthcare is the AI scribe. Doctors are using it when they see a patient to transcribe the visit — and it’s fundamentally a software product that should sit underneath the ONC, which has experience regulating these products,” Farren remarked.

Ambient scribes are just one of the many AI tools being implemented into providers’ software systems, he added. For example, providers are adopting AI models to improve clinical decision making, flag medication errors and streamline coding.

Call for technical standards

Leigh Burchell, chair of the EHR Association and vice president of policy and public affairs at Altera Digital Health, views the plan as largely positive, particularly its focus on innovation and its acknowledgement of the need for technical standards.

Technical data standards — such as those developed by organizations like HL7 and overseen by National Institute of Standards and Technology (NIST) — ensure that healthcare’s software systems can exchange and interpret data consistently and accurately. These standards allow AI tools to more easily integrate with the EHR, as well as use clinical data in a way that is useful for providers, Burchell said.

“We do need standards. Technology in healthcare is complex, and it’s about exchanging information in ways that it can be consumed easily on the other end — and so that it can be acted on. That takes standards,” she declared.

Without standards, AI systems risk miscommunication and poor performance across different settings, Burchell added.

Little regard for patient consent

Burchell also raised concerns that the AI action plan doesn’t adequately address patient consent — particularly whether patients have a say in how their data is used or shared for AI purposes. 

“We’ve seen states pass laws about how AI should be regulated. Where should there be transparency? Where should there be information about the training data that was used? Should patients be notified when AI is used in their diagnostic process or in their treatment determination? This doesn’t really address that,” she explained.

Actually, the plan suggests that the federal government could, in the future, withhold funds from states that pass regulations that get in the way of AI innovation, Burchell pointed out.

But without clear federal rules, states must fill the gap with their own AI laws — which creates a fragmented, burdensome landscape, she noted. To solve this problem, she called for a coherent federal framework to provide more consistent guardrails on issues like transparency and patient consent.

While the White House’s AI action plan lays the groundwork for faster innovation, Burchell and other experts agree it must be accompanied by stronger safeguards to ensure the responsible and equitable use of AI in healthcare.

Credit: MR.Cole_Photographer, Getty Images