
Back in March, after the HHS Request for Information on clinical AI and the ONC HTI-5 proposed rule first emerged, we argued that the industry needed to establish an aligned starting point. Because healthcare organizations span a wide spectrum of AI maturity and governance infrastructure, moving forward safely requires interoperability, equitable access, clinician training, and a practical definition of transparency.
Now, ONC has officially announced its final ruling, triggering a new question. How does HTI-5 impact AI?
The ruling scales back requirements for health IT developers by repealing both source attributes — commonly known as model cards — and risk management requirements. Model cards summarize how an AI solution was developed, its intended purpose, and how it was initially and continuously validated, while the risk management requirements ensured developers had clear plans for the identification and mitigation of known and foreseeable risks. Removing this mandate puts the level of transparency and degree of risk oversight squarely into the hands of health IT developers and healthcare organizations.
Rather than viewing deregulation as an invitation to pull back, we see it as confirmation of MEDITECH’s approach to AI development and our commitment to developer-deployer partnerships.
Evolving the model card
To understand where healthcare AI is heading, we must examine why ONC altered course. Regulators cited a lack of evidence that frontline clinicians actively used technical source attributes at the point of care, projecting over $1.5 billion in regulatory burden reduction by eliminating the mandate. Still, ONC continues to encourage healthcare organizations to request this information from developers.
It isn’t surprising that end users weren’t accessing source attributes. That wasn’t a lack of interest; rather, it was an issue with how to structure information to meet the needs of three distinct groups:
-
Organizational leaders selecting and vetting an AI tool
-
Frontline users employing AI within their workflow
-
Patients engaging with that healthcare organization
Furthermore, a key section in the source attributes and on current model card templates is performance evaluation metrics. Yet, the industry is still trying to figure out which metrics serve as ideal benchmarks for evaluating model and AI system performance. Until we can agree on that, there’s little opportunity for truly comparative evaluation of AI tools. Real transparency requires developers to be honest enough to state when a meaningful benchmark does not yet exist, rather than concealing uncertainty behind aggregated technical statistics.
Acting as information stewards
To solve this, AI developers have a dual obligation as both a technology vendor and a technology resource. In other words, we are information stewards for the solutions we build, guiding customers to meaningful understanding of the information and ensuring organizations come away with clear answers to questions like:
-
What specific clinical or operational problem are you attempting to solve? Is an AI-enabled solution needed, or could the problem be solved through greater optimization of existing technology?
-
What baseline data, technical infrastructure, and workflow maturity does your organization already have?
-
What gaps in understanding must be addressed before this technology is deployed?
-
How will frontline staff interact with the tool, and what oversight mechanisms will govern its output?
Our responsibility is to meet healthcare organizations where they are, helping them evaluate what an AI solution means for their specific operational reality and empowering them to deploy innovation with confidence. And that applies to all organizations across the healthcare landscape, regardless of resources and governance maturity. With a recent Gallup survey finding that nearly 90% of healthcare providers (PCPs) are already using AI tools while only 20% have formal training, closing that knowledge gap is urgent.
Collective responsibility and patient trust
Taking on this stewardship role is critical because innovation evolves so rapidly that it outpaces regulations. So while the nature of AI necessitates guardrails to ensure safe development and deployment, it is increasingly apparent that we can’t rely on creating the perfect standard regulations that can account for every possible scenario or future iteration of the technology. Instead, the health IT industry should focus on the culture that we want to foster.
At MEDITECH, transparency and thoughtful risk management are a baseline commitment, not merely a regulatory checkbox. Where others’ focus has been more on sheer quantity and novelty, our experience has taught us to balance innovation with discipline.
That discipline matters because trust is rightfully fragile right now. According to a CHAI patient survey, 51% of patients say AI would decrease their trust in healthcare, and Wolters Kluwer research shows 75% worry about who is accountable if an AI recommendation goes wrong. If organizations don’t understand the ins and outs of their technology, they can’t monitor it, educate users on when to apply it, or explain to patients why it benefits their care. Setting the right tone starts with developers keeping each other in check and establishing shared rules of the road.
MEDITECH’s commitment to customers
For us, transparency itself isn’t the end goal — meaningful transparency is. Not just access to information, but access to information that healthcare organizations really need, delivered in an understandable format. That requires bidirectional communication and a willingness to actively listen. We offer customers clear channels for direct clinical/IT feedback within the workflow, as well as through user education and onboarding.
We will not only continue to create model cards for our AI solutions, but also continue to have open conversations with customers and others throughout the industry to identify what information and format are most useful. This mirrors our longstanding commitment to interoperability: making data genuinely useful and actionable.
Explore MEDITECH's suite of AI solutions built to support organizations, clinicians, and patients.




