The Medical Group Management Association is calling on federal regulators to establish clear policies for artificial intelligence (AI) in health care that balance innovation with protection for medical practices.
In a Feb. 23 response to the Department of Health and Human Services’ request for information on accelerating AI adoption in clinical care, MGMA outlined priorities that address concerns about liability, transparency and reimbursement facing the nation’s 15,000 medical groups.
The letter responds to an HHS initiative announced in December seeking broad public input on how the department can use its regulatory, reimbursement and research powers to accelerate AI adoption. The request for information, which closed Feb. 23, aims to support what HHS described as a “forward-leaning, industry-supportive, and secure approach” to AI in clinical care.
The association, which represents more than 70,000 medical practice administrators and executives overseeing 350,000 physicians, urged HHS to ensure federal AI policies establish clear guardrails and sustainable payment pathways without adding to administrative burden.
HHS seeks input on AI acceleration
HHS announced the request for information Dec. 19, seeking experience-based feedback from those building, buying, evaluating and using AI tools in clinical care. The initiative, led by the Office of the Deputy Secretary and the Assistant Secretary for Technology Policy/Office of the National Coordinator for Health Information Technology, builds on the administration’s AI policy framework and the department’s internal AI strategy.
In announcing the RFI, Thomas Keane, M.D., assistant secretary for technology policy and national coordinator for health IT, emphasized the importance of data interoperability. “Artificial intelligence is powered by data,” Keane said. “Data liquidity and the trust patients and providers have in how data moves are essential.”
The RFI focuses on three areas: how digital health and software regulatory frameworks should evolve while maintaining patient safety; how reimbursement structures can be simplified to support efficient technologies; and how research and development investments can strengthen implementation science, especially for complex clinical scenarios.
HHS encouraged stakeholders to focus on medium- and long-term trends, including addressing conditions expected to increase in prevalence, such as frailty and dementia.
Mixed results for early adopters
What practices should do now
While MGMA works with HHS on federal policy, practice administrators can take concrete steps to prepare for responsible AI adoption:
Start with governance, even if lightweight. With 56% of medical groups lacking AI governance structures, establishing even basic oversight is critical. This doesn’t require elaborate committees. Begin with a simple intake process for evaluating AI tools and designate someone to track implementations and monitor performance.
Villella offered specific guidance: “Sit down as a team and work on written policies related to acceptable use, safe use. Don’t assume, because it’s so nascent and evolving, that every clinician coming in has the same understanding.” She recommended practices explicitly document current policies and “revisit that because, as you might not actually have AI solutions integrated into your current workflow now, make a plan to evaluate with your vendors.”
Focus on workflow integration first. Before purchasing AI tools, map existing workflows and identify specific pain points. The 44% of practices seeing no workload reduction from AI likely deployed tools without adequate workflow analysis. Implementation success depends on embedding technology into daily operations, not adding parallel systems.
“Instead of a be-all, end-all, like AI is a solution, AI is a feature that is augmenting a solution and making it even more powerful for you in your everyday practice,” Villella said. She recommended practices think about how AI features will be “integrated into trusted, evidence-based, secure solution workflow solutions that we already would be implementing.”
Demand transparency from vendors. Ask specific questions about training data, validation methods and performance metrics. Request information about how tools handle edge cases and what human oversight is required. If vendors can’t or won’t provide clear answers, that’s a red flag.
Account for total cost of ownership. Beyond licensing fees, factor in training time, workflow adjustments, IT infrastructure upgrades and ongoing maintenance. Small practices especially need realistic cost projections before committing to AI tools.
Verify interoperability. Ensure any AI tool can exchange data with existing EHR and practice management systems. Siloed tools that require manual data entry defeat the purpose of automation.
Plan for liability. Review professional liability policies to understand coverage for AI-enabled clinical decision support. Document clinical decision-making processes that involve AI tools, and maintain human oversight of all AI-generated recommendations.
Monitor performance continuously. Don’t assume AI tools will perform as advertised in real-world settings. Establish metrics to track whether tools actually reduce burden, improve documentation or enhance patient care. Be prepared to adjust or discontinue tools that don’t deliver value.
“AI, humans aren’t perfect. We all make mistakes,“ Villella cautioned. “If you’re going to leverage AI, let’s say, to transcribe a lot of ambient listening, transcription, read it through again, make sure there isn’t a mistake.“ She emphasized that AI should be an aid, not a replacement: “It’s not doing your whole documentation job for you. It’s an aid.“
Stay informed on policy developments. As HHS develops AI policies in response to the RFI, practices should monitor how regulatory, reimbursement and research frameworks evolve. MGMA and other professional associations will continue advocating for practice-friendly policies, but administrators should track developments that could affect their specific situations.
According to MGMA polling data, 68% of medical groups reported adding or expanding AI use in 2025, primarily for documentation, patient communications, revenue cycle functions and analytics.
However, adoption hasn’t consistently reduced workload. Among practices using AI, only 39% reported reduced staff workload, while 44% reported no reduction and 17% were unsure.
The documentation challenge is particularly acute. “The friction is this isn’t the reason you do the job. This is just a necessary part of the job,” said Kelly Villella, director of medical education and medical practice at Wolters Kluwer Health, whose company recently surveyed physician associates about AI use. She noted that clinicians want to focus on patient care, but documentation requirements pull them away from that core mission. “How can we make it more efficient to keep that documentation? I think that’s the opportunity for AI.”
For practices that haven’t adopted AI, cost concerns, uncertainty about productivity gains, EHR incompatibility and learning curves present significant barriers, particularly for smaller practices.
Core principles for federal policy
MGMA’s letter outlined five priorities: AI should augment rather than replace clinical judgment, especially in prior authorization decisions; transparency about how tools are designed and validated is essential; workflow integration is critical to avoid increasing burden; oversight should match risk levels; and practices need liability protection for externally developed technology.
“The disruptive part is practices have to change midstream,” said Anders Gilberg, MGMA’s senior vice president of government affairs, describing the broader challenges medical groups face with policy uncertainty. While he was referring to telehealth flexibilities in a recent interview, the observation applies equally to AI adoption, where practices need stable regulatory frameworks to make implementation decisions.
The association expressed concern about recent deregulatory proposals in HHS’s HTI-5 rule that would affect transparency requirements for AI-enabled decision support tools. MGMA urged HHS to maintain comparable transparency mechanisms if these requirements are modified.
MGMA also called for payer transparency, urging HHS to require insurers to disclose their use of AI for utilization management and claims processing. As practices navigate complex credentialing systems, the association emphasized that payer AI systems should be evidence-based and should not interfere with physician clinical decision-making.
Payment reform needed
The letter identified reimbursement as central to responsible AI adoption, urging HHS to explore sustainable payment pathways, grants and demonstration models that recognize AI-enabled efficiencies.
MGMA called for policies accounting for implementation costs and financing challenges faced by small and independent practices. For applications such as ambient documentation and workflow automation, payment models should evolve to reflect value.
The financial pressures on practices extend beyond AI. “The hard part for practices is the transition,” Gilberg said, discussing value-based care models in a recent interview. “Practices often need a cushion, some subsidy” to move into new payment models “without taking an immediate hit to revenue.” The same principle applies to AI adoption, where up-front costs can be substantial before efficiency gains materialize.
Governance gaps remain
MGMA polling from January found 42% of medical group leaders have AI governance structures or policies in place or in development, while 56% have none. This represents progress since late 2024, when nearly three-quarters lacked formal governance despite expanding AI use.
Many practices are deploying AI without adequate oversight frameworks, the data suggest. MGMA encouraged HHS to support scalable governance approaches, including “lightweight” options for diverse practice sizes.
The association noted that the National Institute of Standards and Technology has recognized that organizations must establish governance processes tailored to specific use cases.
The governance gap extends beyond formal structures to day-to-day acceptable use policies. Villella noted in a recent interview that many clinicians may not understand terms like “shadow AI,” the use of unauthorized AI tools in clinical settings.
“I think people don’t even all know the term shadow AI,” Villella said. “You might think, because you use it day to day now in your personal life, as a new clinician, ‘Oh, well, let me run my notes through this.’ That is unacceptable.”
Research priorities
Responding to HHS’s call for input on research and development investments, MGMA urged the department to advance applied research on real-world AI use, noting many medical groups lack resources to independently validate and monitor tools.
The association called for research assessing whether AI reduces administrative burden and integrates effectively into workflows. Studies should examine infrastructure requirements, including cybersecurity capacity and implementation costs.
MGMA noted that performance in controlled environments may not translate to clinical practice improvements, echoing concerns raised in the HHS RFI about the need for real-world evaluation. The association recommended HHS examine AI use in prior authorization with attention to patient access and provider burden, aligning with the department’s interest in reducing health care costs.
Liability concerns
Practice leaders have raised questions about responsibility when AI-enabled tools contribute to patient harm. The letter’s emphasis on protecting practices from liability for externally developed technology reflects these concerns as malpractice considerations evolve.
Broader response to RFI
MGMA’s response represents one of many submissions HHS received on the RFI. The American Hospital Association also submitted comments emphasizing the need to synchronize AI policies with existing regulatory frameworks and ensure appropriate reimbursement for the full spectrum of AI tools.
The Small Business Administration’s Office of Advocacy held roundtables in January to gather feedback from small health care providers, noting barriers including unreliable data quality, lack of education and training, and unresolved challenges related to liability and data governance for AI tools not regulated as medical devices.
MGMA concluded by offering to collaborate with HHS on policies that support innovation while protecting patients and enabling sustainable adoption. For administrators navigating AI implementation, the recommendations offer a blueprint for federal support that could determine whether AI delivers on its promises or adds complexity.
The letter was signed by Gilberg.
