If you saw your doctor in the past year, there’s a reasonable chance your visit was documented by an artificial intelligence system. By some estimates, roughly thirty percent of U.S. physician practices are using AI scribes’ software that listens to your appointment, transcribes the conversation, and drafts your medical note for the physician to sign. The pitch from vendors is clean. AI scribes free physicians from documentation burden. They reduce burnout. They give doctors more eye contact with patients. There are real benefits. There is also money changing hands. Hospital systems sign multi-million-dollar contracts with AI scribe vendors. Physicians and practice executives attend vendor-sponsored conferences. Clinical thought leaders publish on vendor-funded research panels.

If your doctor had accepted a free dinner from a pharmaceutical company last year, you could go to a federal database called Open Payments and find the receipt. You could see the company name, the dollar amount, the date. That database exists because Congress decided, in 2010, that patients had a right to know which financial relationships their physicians were in. But if your doctor’s hospital signed a multi-million-dollar contract with the AI vendor whose software just wrote your visit note a contract that influences how your care is documented, billed, and tracked for years to come there is no equivalent database. The money is real. The relationships shape clinical practice. The disclosure does not exist. This is the disclosure gap. And it has health equity implications most patients have never been told about.

What Open Payments Actually Covers

In 2010, as part of the Affordable Care Act, Congress passed the Physician Payments Sunshine Act. The law requires pharmaceutical companies, medical device manufacturers, and biologics manufacturers to publicly report payments they make to physicians and teaching hospitals.

The result is the Open Payments database, maintained by the Centers for Medicare and Medicaid Services. According to CMS, the database contains more than fourteen million payment records covering reporting years 2018through 2024. Anyone with internet access can search it by physician name, company, state, specialty, or payment type. The threshold is low any payment above ten dollars per instance, or one hundred dollars in aggregate per year, must be reported.

The payments covered run the spectrum: consulting fees, speaker honoraria, travel reimbursements, meals, research grants, royalties, gifts, and ownership interests. According to a 2021 analysis of 2014-2018 Open Payments data published in PLOS ONE, more than fifty million payment records were issued to over 770,000 physicians during that period, totaling over $8.7 billion transferred from industry to physicians.

The premise behind the Sunshine Act was simple: when industry pays providers, the relationship creates the potential for conflict of interest. Patients deserve to know. Researchers deserve to study the patterns. Journalists deserve to ask questions. The database is the receipt.

What the Sunshine Act does not cover is just as important.

The Disclosure Gap

The Sunshine Act’s reporting requirements apply to a specific category of company: ‘applicable manufacturers’ of ‘covered drugs, devices, biologicals, or medical supplies’ that are reimbursable under federal healthcare programs like Medicare or Medicaid. The definition is precise. It was written in 2010 when the pharmaceutical and medical device industries were the dominant concerns. The Act was not written with software in mind.

Today, that gap matters more than ever. A modern healthcare AI vendor may sell an ambient documentation tool that listens to clinical conversations and drafts notes. It may sell a clinical decision support tool that recommends diagnoses or treatment options. It may sell a risk prediction algorithm that flags which patients to monitor more closely. These tools shape clinical practice. They are not pharmaceutical compounds. They are not implantable devices. They are software.

Some healthcare AI products do fall under FDA regulation as ‘software as a medical device’ the FDA finalized guidance on its Predetermined Change Control Plan framework for AI/ML-enabled medical devices in December 2024. When an AI product is FDA-cleared as a medical device, and its manufacturer makes payments to physicians, those payments are reportable under Open Payments. But the majority of clinical AI tools currently deployed in U.S. hospitals and physician practices are not classified as medical devices. They are classified as administrative or workflow tools. That single classification distinction medical device versus administrative tool determines whether the financial relationship between the AI vendor and your physician shows up in a federal database, or stays invisible.

When an AI scribe drafts your medical note, the AI vendor’s financial relationship with your physician’s hospital is usually invisible to you, your family, and the journalists who would normally be checking these relationships on your behalf.

It is worth pausing here to note what isn’t the issue. The issue isn’t that AI scribes exist. Many physicians find them genuinely useful. The issue isn’t that hospitals contract with software vendors. Hospitals contract with hundreds of software vendors. The issue is that one specific category of vendor relationship one with direct downstream consequences for clinical documentation, billing, and patient records falls outside the disclosure framework that Congress put in place precisely to give patients transparency about industry-provider relationships.

Why This Disclosure Gap Matters

The Sunshine Act exists because researchers and policy advocates spent years documenting the ways that industry payments correlate with prescribing behavior. The evidence is not subtle. A 2023 study published on medRxiv analyzed industry payments to anesthesiologists between 2014 and 2022 using Open Payments data. The researchers found, consistent with prior literature, that non-research

payments to physicians from opioid manufacturers were significantly associated with increased opioid prescriptions at the individual physician level, and with higher opioid overdose deaths at the county level. These findings were not isolated. A 2024 ScienceDirect review of the past decade of Open Payments literature found consistent patterns across multiple specialties: physicians who received payments from a pharmaceutical manufacturer were more likely to prescribe that manufacturer’s products, often at rates that could not be explained by clinical guidelines alone.

This is the foundational research that makes Open Payments meaningful. The database matters because the relationships matter they show up in clinical decisions, in prescribing patterns, in patient outcomes. Now consider what the equivalent question looks like for clinical AI tools. A hospital system contracts with an AI scribe vendor. The vendor’s software writes the clinical notes for thousands of patient encounters every month. Those notes inform billing codes. Those codes influence reimbursement.

The vendor may also offer clinical decision support that recommends diagnostic workups or treatment options. If a financial relationship between the hospital and the vendor shapes how the software performs which diagnostic recommendations it surfaces, which billing codes it suggests, which clinical patterns it flags there is no public database where patients, journalists, or researchers can verify what relationship exists.

What This Has to Do With Black Patients

Here is where the disclosure gap meets health equity and where, in our

view at Melanin Bliss Media, this story has to be told. Research has documented racial disparities in how AI tools perform across patient populations. A 2021 study by researchers at the University of Chicago Booth School of Business, published in Nature Medicine, demonstrated that an algorithm trained on patient knee X-rays could identify sources of pain in Black patients that human radiologists had been missing for decades. The implication was profound: existing clinical assessment had been systematically underestimating Black patients’ pain.

The algorithm, designed differently, could correct for the bias. Stanford’s Institute for Human-Centered AI has documented similar findings: that the design choices made by clinical algorithm developers which training data they use, which outcome variables they optimize for, which validation populations they test on directly determine whether the resulting tools reduce or entrench racial disparities in care.

Generative AI models, including the type that power many clinical AI scribes, show similar patterns. Research at Mass General Brigham has demonstrated that large language models can effectively surface social determinants of health from clinical notes a potentially powerful tool for documenting the housing, food access, and transportation barriers that disproportionately affect Black and Brown patients. But the same research has shown that performance varies across patient demographics, and that systematic auditing is required to ensure the tools serve everyone equally.

If the financial relationships between hospital systems and AI vendors are invisible to the public, the equity assessments are functionally invisible too. The patients most likely to be harmed by biased AI tools are the patients least likely to know which tools their hospital deployed, who profits from them, or whether they were ever tested for fairness.

When the HHS Office for Civil Rights finalized its Section 1557 nondiscrimination rule in May 2024, it included provisions specifically addressing the use of patient care decision support tools. Covered entities most U.S. hospitals are now legally required to make reasonable efforts to identify and mitigate discrimination risks in the patient care decision support tools they use. The Section 1557 rule is part of a broader regulatory current. The Coalition for Health AI released its Responsible Health AI Framework in 2024. Joint Commission announced a new Responsible Use of Health Data certification program in 2024, in partnership with CHAI. The direction of travel is unmistakable: regulators are catching up to clinical AI.

But none of these frameworks require disclosure of the financial relationships between hospital systems and the AI vendors they contract with. The equity assessments can be required without the transparency that would let patients understand whose interests shaped the assessments in the first place.

Why This Should Worry Every Patient

There is a version of this story that says: this is a technical regulatory gap, it will be closed in time, the existing framework is doing its best. There is another version that says: this gap is the predictable consequence of a regulatory system that catches up to industry practice years after the practice is established, and the patients who pay the price are disproportionately Black, Brown, low-income, and uninsured.

Both versions are true. The first describes the mechanics. The second describes the cost. We have seen this pattern before. When direct-to-consumer pharmaceutical advertising expanded in the 1990s, the disclosure frameworks took years to catch up and the patients who paid the price were the ones with the least

information and the least access to alternative care. When opioid prescribing patterns surged in the 2000s and 2010s, the documentation of payment-prescribing correlations came too late for the communities hardest hit by overdose. Each time, the disclosure framework lagged the industry practice.

Each time, marginalized communities bore the costs of the lag. Clinical AI is not pharmaceutical compounds. It is not exactly the same shape of risk. But the underlying pattern industry practice racing ahead of disclosure frameworks, with predictable downstream consequences for marginalized patients is the same shape.

What Closing the Gap Looks Like

Closing the disclosure gap is a policy question that does not yet have an easy answer. The Sunshine Act’s definitions would need to be updated,

either by Congress or by CMS regulatory action. The categories of ‘applicable manufacturers’ and ‘covered products’ would need to evolve to include the software vendors whose products now shape clinical practice. Threshold rules would need to be adapted for software licensing arrangements that don’t map neatly to the per-instance payments the current Sunshine Act covers.

None of this is impossible. The Sunshine Act has been amended before. In 2021, reporting requirements expanded to include payments made to advanced practice registered nurses and physician assistants. The framework can evolve. The question is whether the political will exists to update it for the AI era, and whether patient advocates and journalists demand the update loudly enough.

In the meantime, hospital systems can choose to be transparent voluntarily. Some already publish their major vendor relationships through community benefit reports or annual reports. Most do not. Patient advocacy organizations can press for disclosure as a condition of nonprofit tax-exempt status review. Foundations funding clinical AI deployment in safety-net settings can require disclosure as a grant condition. Federal grantmakers NIH, ARPA-H can require transparency in any AI-enabled clinical research they fund. None of these moves require new legislation. All of them depend on patient advocates, journalists, and policy entrepreneurs asking the right questions, loudly.

What Readers Can Do

If you are a patient: when you receive care, you have the right to ask your physician or your hospital what AI tools they use in your care. You can ask whether those tools have been audited for racial or demographic disparities. You can ask whether your medical note was drafted by an AI scribe and whether the documentation accurately reflects what you said. The questions are not rude. They are appropriate.

If you are a healthcare professional working inside a hospital system: you can ask, internally, what your institution’s AI vendor contracts include and whether equity audits have been conducted. You can advocate for transparency frameworks in your own institution. Many of the improvements in healthcare disclosure over the past two decades came from inside the system, not outside.

If you are a policymaker, journalist, or advocate: this is a story that needs

more reporting than has been done. The Open Payments database remains a remarkable public resource, but it is an artifact of the disclosure logic of 2010. The technology that shapes clinical practice in 2026 has outpaced that logic. The patients most affected by the gap deserve someone to close it.

At Melanin Bliss Media, we will keep reporting on this. The next phase of our coverage will look at specific hospital systems’ AI vendor relationships, the equity audit practices of major clinical AI vendors, and the policy advocates working to update the Sunshine Act for the software era. If you have information that should be reported, or if you are working inside an institution that deserves coverage on this issue, we welcome confidential outreach at info@mbmedia.co.

Public health, reimagined. Not a brand. A blueprint.

EDITORIAL DISCLOSURE

MB EDITORIAL FIREWALL DISCLOSURE: Melanin Bliss Media operates two business lines under one LLC MB Media (this journalism platform) and MB Strategy Consulting AI (clinical AI advisory services). This investigation discusses healthcare AI vendor practices generally and does not name specific MB Strategy Consulting AI clients. MB Strategy Consulting AI does not represent any healthcare AI vendor. MB editorial coverage is not influenced by MB Strategy Consulting AI engagements. The firewall is the line. We hold it.

Sources Cited in This Piece

• CMS Open Payments database (openpaymentsdata.cms.gov)

• Physician Payments Sunshine Act of 2010 (Affordable Care Act, Section 6002)

• PLOS ONE: ‘Physicians payment in the United States between 2014 and 2018: An

analysis of the CMS Open Payments database’ (2021, PMC8171935)

• MedRxiv: ‘Industry payments to anesthesiologists in the United States between 2014

and 2022’

• ScienceDirect: Overview of Open Payments, advanced practice registered nurses

public reporting (2024)

• FDA: Predetermined Change Control Plan framework for AI/ML-enabled medical

devices (December 2024)

• HHS Office for Civil Rights: Section 1557 final rule, patient care decision support

tool provisions (May 2024)

• Coalition for Health AI (CHAI): Responsible Health AI Framework (2024)

• Joint Commission: Responsible Use of Health Data certification program (announced

2024)

• University of Chicago Booth: ‘Reducing Racial Disparities in Knee Pain using

Artificial Intelligence’

• Stanford Institute for Human-Centered AI (HAI): research on clinical algorithm bias

• Mass General Brigham: Generative AI Models Effectively Highlight Social

Determinants of Health in Doctors’ Notes (2024 research findings)