June 22, 2026

U.S. Representative Jay Obernolte
2433 Rayburn House Office Building
Washington, DC  20515

U.S. Representative Lori Trahan
2233 Rayburn House Office Building
Washington, DC 20515

Re: Great American AI Act Discussion Draft

Submitted electronically via GAAIA@mail.house.gov

 

Dear Representatives Obernolte and Trahan:

The Workgroup for Electronic Data Interchange (WEDI) writes today in response to the “Great American Artificial Intelligence Act (GAAIA) Discussion Draft” released on June 4, 2026. We are pleased to submit comments on what could be significant legislation for artificial intelligence (AI), and the innovation and role it is playing in the health care industry.

WEDI was formed in 1991 by then Department of Health and Human Services (HHS) Secretary Dr. Louis Sullivan to identify opportunities to improve the efficiency of health data exchange. Named in the Health Insurance Portability and Accountability Act (HIPAA) legislation as an advisor to the Secretary of HHS, WEDI is the leading multi-stakeholder authority on the use of health information technology (IT) to efficiently improve health information exchange, enhance care quality, and reduce costs. With a focus on advancing standards for electronic administrative transactions, and promoting data privacy and security, WEDI is recognized and trusted as a formal advisor to the Secretary. Our diverse membership includes health plans, providers, standards development organizations (SDOs), health IT vendors, clearinghouses, federal and state government agencies, and patient advocacy organizations.

WEDI’s initial reaction to this draft bill was that including the Secretary of Health and Human Services (HHS) and the health care industry legislation would significantly improve the impact of the legislation. AI is rapidly becoming one of the most transformative technologies in health care, offering significant opportunities to improve patient care, enhance operational efficiency, and reduce administrative burden. AI-powered tools are being used to support clinical decision-making, identify patterns in large volumes of health data, improve diagnostic accuracy, predict patient risks, and personalize treatment plans. Beyond clinical applications, AI is automating routine administrative tasks, such as claims processing, prior authorization workflows, customer service, and data management, allowing clinicians to focus more time on patient care. As health care organizations continue to face increasing demands related to cost, workforce shortages, interoperability, and consumer expectations, AI has emerged as a critical catalyst of innovation, helping organizations leverage data more effectively while supporting better outcomes, improved experiences, and a more efficient health care system. WEDI strongly recommends that health care be a focus of this bill and the HHS Secretary be explicitly included as one of the Secretaries to be consulted in the various actions called for in the bill.

Background

AI is currently deployed by providers and health plans and the vendors that support them. The increased exchange and use of electronic health data will increase use of AI in health care. Currently, the most common applications of AI in health care include diagnosis and treatment recommendations, patient communication, engagement and adherence, and administrative activities. Although there are some examples where AI may perform health care tasks as well or better than humans, there remain critical implementation issues that need to be addressed prior to wide-scale deployment of this technology.

As AI and related technologies become increasingly established in the health care sector, it is important to recognize that AI in health care is not a single technology, but rather several. These include:

• Machine learning (ML). ML is a statistical technique for fitting models to data. These models ‘learn’ through data training and are one of the most common forms of AI.
• Deep learning and neural networks. The most complex forms of ML involve deep learning or neural network models with many levels of features or variables that predict outcomes. The “deep” in deep learning refers to the depth of layers in a neural network. A neural network that consists of more than three layers, which would be inclusive of the inputs and the output, can be considered a deep learning algorithm.
• Natural language processing (NLP). NLP includes applications such as speech recognition, text analysis, translation, and other goals related to language. There are two basic approaches: statistical and semantic NLP. Statistical NLP is based on ML (deep learning neural networks in particular) and has contributed to a recent increase in accuracy of recognition. Semantic analysis analyzes the grammatical format of sentences, including the arrangement of words, phrases, and clauses, to determine relationships between independent terms in a specific context.

AI systems can scrape, search, aggregate, and learn from data from the internet about an individual (voice from videos, financial accounts, likes on social media or shopping sites, etc.) and use that data for both positive and negative purposes. The more robust the data source the more robust the AI technology grows (and learns) and the potentially more dangerous this could become. For example, AI could use a Facebook post about a person being in the hospital to triangulate a specific clinician in the area, locate the individual's address and send them a convincing, but fraudulent, bill requesting payment. At risk is protected health information and financial information as passwords and secret information as safeguards and validation methods become easier to compromise because of AI. The rapid emergence of AI in health care has been revolutionary, redesigning the way patients are diagnosed, treated, and monitored. This technology is drastically improving health care revenue-cycle processes, advancing research and outcomes, producing more accurate diagnoses, and enabling more personalized treatments.

Comments

Our comments will address five critical issues related to the use of AI in health care:

1. 1. AI and the patient perspective
   2. Opportunities that AI offers health care providers and health plans
   3. The challenges associated with the data leveraged for AI – the potential for bias
   4. Other challenges AI presents for health care stakeholders
   5. Recommendations for federal government action

1. AI and the patient perspective

From the patient’s perspective, the goals of using AI in the delivery of health care services are to improve outcomes and reduce overall cost in the health care system and direct costs to the consumer. Health care is already expensive on a macro level, as well as an individual one for many consumers and many fear that new AI-based technologies will drive up costs.

At the same time, AI systems can be used to inform the patient regarding the availability of providers and the expected cost related to medical services. AI can improve provider directories and accurately signal to patients if a provider is accepting new patients. With cost transparency being an important issue, patients can leverage AI to give them an accurate estimate of the costs of a medical service, treatment, medication, or device. Plans are also looking at using AI in their cost calculators by using historical claims data of what services cost.

Patients are also concerned about the potential loss of that personal connection with their clinician. Given the choice, patients envision AI systems “helping the clinician come up with better treatments for me” is far preferred to AI systems “getting between me and the clinician.” There are already issues with electronic health records (EHRs) and computerized clinical encounters and the potential for less personal interactions. It is important to remember that many patients who are seeking health care are not feeling well and are at their most vulnerable. When patients are seeking care, they really value relationships and trust. Patients fear that high tech will replace high touch.

2. Opportunities that AI offers health care providers and health plans

AI has the potential to assist clinicians and improve patient care. In health care, the dominant applications of NLP involve the creation, understanding and classification of clinical documentation, and published research. NLP systems can analyze unstructured clinical notes on patients, prepare reports (e.g., on radiology examinations), transcribe patient interactions, and conduct conversational AI. Improvements to clinical care can be accomplished through:

• Clinical decision making (Computer Decision Support)
• Assisting in making real-time inferences for health, risk, alert, and health outcomes prediction
• Closing gaps in care through analysis of claims data
• Early prediction of disease for patients and populations
• Retrospective analysis of disease through analysis of data sets
• Extracting information from large patient populations
• Learning and self-correcting abilities to improve its accuracy based on feedback
• Augmenting or even replacing human judgment in certain functional areas, including automating detection of relevant findings for pathology, radiology, and other medical specialties
• Reduction of medication errors and adverse events
• Modeling and preventing the spread of hospital acquired infections
• Identification of patient subgroups for personalized and precision medicine
• Robot-assisted surgery

Provider business functions can also be improved through the appropriate use of AI in the following areas:

• Optimizing hospital processes, such as resource allocation and patient flow
• AI computer-assisted clinician documentation
• Simplifying administrative workflow, e.g., letter and appeals writing, scribing services, improve EHR usability
• Improving patient communications and consumer services through auto-generated letters and chat bots
• Identifying billing errors prior to submission
• Reducing burdens associated with prior authorization processes
• Improving quality improvement reporting and researching large sets of data over a long period of time looking for patterns for improvement
• Offering real-time education opportunities for clinicians as they provide care
• Supporting value-based care programs-supplying clinicians with real-time, actionable data
• Streamlining credentialing, provider directories, privileging, and other database-related initiatives
• Ultimately, reducing clinician burnout

For health plans, business functions also have the potential of being improved through the leveraging of AI. These improvements could include:

• Supporting real-time identification for providers of lower cost options, e.g., optimum site of service
• Improving claims denial process
• Improving prior authorization processes (Health plans are exploring leveraging large language models to streamline the electronic prior authorization communication between plans and providers.)
• Using data and AI for research to look at disease progression
• Improving fraud detection
• Enhancing cybersecurity protections

3. The challenges associated with the data leveraged for AI – the potential for bias

Bias in AI has the potential of causing harm and must also account for human and systemic biases. Human biases can relate to how people use data to fill in missing information, such as assumptions about a person based on their neighborhood of residence. Systemic biases result from institutions operating in ways that disadvantage certain social groups. Harm is possible when human, systemic, and computational biases combine. Researchers at the National Institute of Standards and Technology (NIST) have recommended widening the scope for the source of these biases to broader societal factors that influence how technology is developed.

Biases can be embedded in and related to AI, and that has the potential to perpetuate disparities in health care, causing harm when that bias is, for example, treatment support recommendations or prior authorization. Some have suggested that the end user (patient and clinician) be provided with a type of “nutrition card” to inform them of the underlying data used in the AI engine and the potential for bias. Another perspective is for the software developer to have a “recipe card” to help understand the implications of the data.

Clinicians need to understand the AI being used and how the data is driving the decision-making. It is critical to have a transparent data source or sources, so clinicians know what the data was trained on. “Black box” algorithms, AI systems that are not transparent with their data sources and how they get to their decision-making endpoints, create a lack of trust with clinicians and patients alike.

We also recognize that AI systems learn as they go and learn over time. The data originally used, however, can become stale or the population it serves could evolve. A process should be established to recalibrate the data and the AI and the targets to ensure those changes are factored in and current. Data is not only the source of AI calculations and outputs but also its errors and biases.

We also understand that bias is inherently a human problem. The industry should explore how to use “good” algorithms and “good” AI tools to identify and address systemic bias in an AI engine. Creating anti-bias AI-algorithms that identify when an algorithm itself is biased would also be a step in the right direction.

4. Other challenges AI presents for health care stakeholders

AI presents other challenges to the health care sector, including the following:

• High initial capital requirement for providers and health plans
• AI “trust” factor (patients and clinicians)
• Integration issues and ensuring that AI systems work seamlessly with existing health IT systems and processes
• Lack of Interoperability between AI solutions
• Ethical issues related to the loss of empathy, kindness, and appropriate behavior when dealing with AI systems that do not possess compassion
• Reluctance among clinicians to adopt
• Fear of replacing humans and the potential for increased unemployment
• Clinicians may encounter patients that use AI tools on their own and have unreal expectations about their diagnosis and treatment
• Concern regarding patient data privacy and security
• Need for continuous training by data from clinical studies and other data sources
• Lack of incentives for the entities that compile and exchange information that inform AI systems
• Additional burden on clinicians to learn something new and a new process to factor into their decision-making
• Lack of curated health care data to develop effective AI
• Unknown liability issues for clinicians and developers when AI is used by and relied on and there is an adverse outcome
• Lack of regulatory guidelines

5. Recommendations for federal government action

The U.S. is making progress in developing AI regulation, including the NIST “AI Risk Management Framework” and existing laws and regulations that apply to AI. More, however, is needed. AI systems are becoming increasingly powerful and having impact much faster than the federal government can react. With the incredible pace of development and improvement, some question whether the government is even capable of regulating AI effectively.

ONC Activity

The Office of the National Coordinator for Health Information Technology’s (ONC’s) regulation entitled “Health Data, Technology, and Interoperability: Certification Program Updates, Algorithm Transparency, and Information Sharing” (HTI-1) establishes first of its kind transparency requirements for AI and other predictive algorithms that are part of certified health IT. ONC-certified health IT supports the care delivered by more than 96% of hospitals and 78% of office-based physicians nationwide. This regulatory approach aims to promote responsible AI and make it possible for clinicians to access a consistent, baseline set of information about the algorithms they use to support their decision making and to assess such algorithms for fairness, appropriateness, validity, effectiveness, and safety.

While the certification program for health IT, which includes EHR technologies, is voluntary for software developers, hospitals and clinicians must use certified systems when participating in certain Centers for Medicare & Medicaid Services (CMS) payment programs. Thus, the inclusion of AI requirements in software certification is an effective method of reaching the inpatient and outpatient communities.

ONC anticipates that increased transparency will help avoid unintended consequences of algorithmic bias. Specifically, the AI regulations build upon their existing certification requirements for clinical decision support systems by defining a new category for predictive tools, which includes AI and algorithms. ONC’s approach is to not prescribe or prohibit any particular algorithm but rather increase transparency as a way to help people “navigate” the technology. We concur that the more information about the data sets AI systems are trained on would boost transparency and increase the level of trust clinicians have in the technology.

Additional government action

We know that AI tools have raised concerns in the health care industry for their potential to make mistakes, spread misinformation, and perpetuate biases. We also know that while self-regulation may be appropriate for some less critical AI use cases, other more critical uses of AI require the government to closely monitor behavior and enforce non-compliance of established regulations. A challenge, however, for policymakers is simply knowing what questions to ask. As this technology and how it is applied to health care is still new, the industry is continuing to grapple with how best to regulate it.

WEDI believes the following tenets should underpin federal government action in health care AI:

• Encourage innovation and research. While we agree that government oversight over the use of AI in health care is imperative, that oversight should not stifle innovation through research and discourage the development of new and enhanced AI use cases.
• Ensure transparency. For patients, clinicians, and others that leverage AI tools, it is critical that each understand how the AI system was developed, what data was used in the creation of the tool, and be informed of any expected biases.
• Maintain accountability. AI vendors and those health care entities that use AI must be accountable for any problems, errors, or other issues that negatively impact the care delivery process.
• Protect the privacy and security of patient data. Health care has the HIPAA privacy and security regulations as guardrails to ensure patient data is appropriately captured, used, and shared. As AI tools are developed and deployed in the health care sector that are not covered under HIPAA, both vendors and end users of AI must be required to adhere to appropriate privacy and security safeguards.

We put forward the following recommended government actions:

1. Creation of an AI oversight structure within the federal government. Coordinating federal government AI policy and activity will be critical. We urge the government to develop an AI oversight structure that includes:
   1. An Administration level position that coordinates AI policy across all branches of the federal government and acts as a central point of contact.
   2. An AI-focused office within HHS to coordinate on health care specific issues.
   3. CMS-specific directions on AI related administrative and clinical issues.
   4. Food and Drug Administration-specific directions related to regulating AI software as or in a medical device.
   5. NIST-specific directions on AI-related privacy and security issues.
   6. Office for Civil Rights-specific directions related to the impact of AI on the privacy and security of protected health information.
   7. ONC-specific directions on the use of AI in health IT.
2. Overall government funding should be increased in the AI space and health care specifically. There is tremendous potential associated with AI in the health care sector. The federal government should identify funding opportunities to advance the appropriate use of AI in health care.
3. Government research into AI and health care should be accelerated. Areas of research could include: 
   1. Whether AI in health care creates added or reduces costs for patients and other stakeholders.
   2. If and how new scientific and clinical findings should be shared through open-source methods.
   3. The need for data sets used in AI models to be reviewed by the government for their clinical applicability, overall clinical value, and ability to decrease costs.
   4. The need for continuous training by data from clinical studies and other research sources.
4. Develop consistent definitions. The federal government should develop AI and AI-related definitions to ensure consistency across all government agencies and to provide clear direction for the private sector. For example, it would be helpful to see established industry-wide or regulatory agreement on a concrete definition of fairness and bias, a uniform standard of how to measure it, and standards on how to report it.
5. Establish collaboration between government and the private sector. AI capabilities are exhibiting exponential growth. It is imperative that the federal government work in tandem with the private sector to ensure that the government is aware of these capabilities and that appropriate policies are developed and implemented. The government should also collaborate with the private sector to evaluate AI models and threats.
6. Draft a consumer AI “bill of rights.” When AI is used to make or contribute to a decision that impacts a consumer, especially in an adverse way, the consumer has a right to know that: a) AI was used and b) how it was used. When consumers have this information, they can make more informed decisions.
7. Consider AI “nutrition labels” to warn patients. Nutrition type labels (aka “warning labels”) would be helpful as they could both educate the patient regarding the use of AI for a service or treatment and warn the patient of potential harm. Further work is needed on when the patient should receive the information and for which services. We recommend working directly with patient advocacy organizations to determine the most appropriate approach to these types of AI warning labels.
8. Require transparency. When AI is used to make or contribute to a decision that impacts a consumer, especially in an adverse way, the consumer has a right to know that AI was used and how it was used. When consumers have this information, they can make informed decisions to object or remediate such decisions. Further, AI is very skilled at presenting information convincingly, even when that information is wrong. Even when an expert reviews the results, they can be misled that it is accurate, especially if the message is nuanced. The consumer is entitled to a human review of the data, analysis, and conclusion when they are potentially disadvantaged by the result. Possible mitigations include approaches such as labels that AI was used. For example, texts that are voice dictated have the disclaimer 'Sent with Siri'. Further, generative foundation models, like GPT, should have to comply with transparency requirements—disclosing, for example that AI generated the content.
9. Classify low-risk, medium-risk, and high-risk AI. While AI can impact health care in many ways, the potential for danger or harm will vary significantly depending on the AI system and in what area of health care it is utilized. We urge the federal government to classify AI in terms of its potential to cause harm to patients or increase burden and cost to end users. Once this classification is in place, it will facilitate the development of supporting public policy.
10. Prioritize the development of AI privacy and security policies. The inappropriate use of AI has the potential of resulting in improper disclosure of protected health information. This in turn could lead to a decreased level of trust in AI by patients and clinicians. The federal government should prioritize the development of supporting privacy and security policies to minimize the chance of improper disclosures and create an enforcement process for those that violate AI privacy or security policies.
11. Establish standards. A governing body could be formed to establish guidelines for AI use and recognizable labels that can be applied to AI or use cases. This will help guarantee the public that the AI in each situation will function according to a set of expectations defined by the label. The certification would prove that the AI met the qualifications and further allow the consumer to identify risks and unintended consequences to their own interests. For example, manufacturers cannot use the term 'low fat' or 'low cholesterol', etc. unless those foods meet the defined criteria of 'low fat' and 'low cholesterol.'
12. Work toward interoperability. HHS has as one of its priorities the implementation of interoperable health IT leading to the more efficient and effective exchange of interoperable health system via FHIR APIs and we encourage the federal government to incorporate AI into its interoperability policy and regulation.
13. Explore the creation of a risk of bias scoring process. AI is only as good as the data it is drawing from. If that data is incomplete, too limited, poor quality, lacks diversity, or is misapplied to a different population type, the consumer could be negatively impacted. Working with impacted stakeholders, we recommend the government explore developing a scoring process that measures how diverse the dataset and weighing that against the population being served.
14. Focus on education. There needs to be a national, coordinated education effort for both patients and providers. This education should be focused on general AI principles, as well as specific applications, models, and algorithms. We urge the federal government to work directly with WEDI and other stakeholder organizations to convey a consistent message on the opportunities and challenges related to AI use in health care.
15. Create an AI complaint process. Both patients and providers should have the ability to lodge complaints regarding how AI was used, and any potential biases or errors associated with that AI system. This could mirror the current process of reporting violations of the HIPAA privacy and security regulations, where individuals and organizations are encouraged to come forward when they believe they have experienced an unauthorized data breach or other violation.
16. Create an adverse event reporting process. In terms of adverse event reporting, we urge a process similar to what is used by the Federal Aviation Commission National Safety Board where all the airlines share adverse event data together and review the issues associated with the adverse event. The process should be to gather the data and be made aware of the issues that led to the error or mistake rather than to be punitive. There should also be established a centralized authority that collects information on when algorithms make mistakes. The overall goal should be to make AI as safe as possible, which can be achieved through an independent third-party repository for errors.
17. Incorporate AI into health IT certification. ONC should include AI in its EHR certification program and any other appropriate certification programs. Although ONC certification is voluntary for software developers, hospitals and clinicians must use certified systems when participating in certain CMS payment programs.
18. Coordinate between different AI solutions. The federal government should seek to coordinate AI systems and ensure AI systems being designed to work with each other to avoid producing discrepant and conflicting results or the kinds of harm to patients.

Conclusion

WEDI applauds your efforts to solicit public feedback on the GAAIA Discussion Draft and potential actions for the federal government to take in promoting effective use of AI and creating appropriate guardrails. AI currently has a significant impact on the health care sector and that impact will continue to grow exponentially, affecting patients, clinicians, health plans, and others. We appreciate the opportunity to share our perspective on this proposed legislation. We hope our comments and recommendations will serve to assist you as you move forward with the bill. Please contact Robert Tennant, WEDI Executive Director, at rtennant@WEDI.org with any questions on these comments and recommendations.

Sincerely,
/s/

Merri-Lee Stine
Chair, WEDI

cc: WEDI Board of Directors