Maryland’s policy on AI in healthcare misses the big picture | GUEST COMMENTARY

Maryland wants to lead the nation in regulating artificial intelligence in healthcare. The question is whether it truly understands the ecosystem it is attempting to regulate.

Last year, Maryland passed House Bill 820, legislation aimed at restricting how insurers, pharmacy benefit managers and utilization review organizations use artificial intelligence in healthcare decision-making. Supporters framed the law as a necessary consumer protection measure designed to prevent automated denials of care and ensure that human clinicians remain

But beneath the political language and legislative optimism lies a far more complicated reality: Modern healthcare AI does not operate in isolation. AI systems are only as reliable as the data they ingest, and increasingly, that data is no longer being generated solely by physicians.

Today, medicine runs on an interconnected digital infrastructure composed of electronic medical records, ambient AI scribes, automated coding tools, transcription platforms, predictive analytics systems and large language models embedded directly into clinical workflows. Most physicians now use some form of AI-assisted documentation. Conversations between patients and doctors are being recorded, summarized, structured, coded and integrated into medical records in real time. These records are being used by insurers to literally eavesdrop on what is happening in the doctor’s office through the AI scribe. How will the state regulate this?

The medical chart is no longer purely a physician-authored document. It is becoming a hybrid product generated jointly by clinicians and machines.

That matters because insurer AI systems reviewing prior authorizations or determining medical necessity depend entirely on the quality of those inputs.

If flawed documentation enters the system upstream, flawed decisions emerge downstream.

Yet Maryland’s law focuses primarily on insurer-side AI while paying comparatively little attention to the broader infrastructure generating the data that those insurers rely upon.

If an AI-driven utilization management system denies care because documentation was incomplete, templated incorrectly, misinterpreted by a scribe or poorly summarized by an automated tool, accountability itself becomes increasingly difficult to define because responsibility is now distributed across multiple technologies, platforms, institutions and actors operating simultaneously within the same clinical workflow.

More importantly, does Maryland itself possess the technical expertise and operational resources necessary to meaningfully evaluate these systems?

The law mandates oversight, auditing, transparency and human review. But what does “audit” actually mean in the age of large language models and adaptive algorithms?

Can state regulators independently evaluate proprietary insurer models?

Can they assess training datasets, bias amplification, prompt architecture or algorithmic drift?

Can they distinguish whether an adverse outcome resulted from flawed physician documentation, flawed AI summarization, flawed data extraction, flawed coding or flawed model reasoning?

Or are we creating the appearance of oversight while regulators remain technologically outmatched by the systems they are regulating?

At the same time Maryland is attempting to regulate insurer AI aggressively, healthcare systems themselves are rapidly deploying AI internally to address physician burnout, staffing shortages, administrative overload and operational inefficiencies.

Hospitals and physicians are already using AI for triage support, radiology prioritization, predictive analytics, retinal imaging analysis, workflow optimization, medication safety review and clinical augmentation.

Ironically, last year, Maryland lawmakers also considered proposals that would have significantly restricted physicians from using AI in clinical decision-making.

That approach fundamentally misunderstands the trajectory of modern medicine.

No responsible physician is arguing that AI should replace medical judgment.

But augmentation is not replacement. A radiologist using AI to identify subtle imaging abnormalities is not surrendering expertise. An ophthalmologist using AI-assisted retinal analysis is not abandoning clinical judgment.

A physician using AI to synthesize large volumes of patient data or flag dangerous medication interactions is not practicing irresponsibly.

In fact, as AI systems mature, failing to use augmentation tools in certain settings may eventually become harder to defend.

The real debate should never have been “AI versus physicians.”

The real issue is whether AI is being implemented transparently, responsibly and accountably — and whether regulators themselves understand these systems well enough to govern them competently.

That is where Maryland faces its greatest challenge.

The state cannot claim technological leadership while operating within healthcare infrastructures that still struggle with interoperability failures, fragmented public health systems, delayed Medicaid processing, workforce shortages and legacy IT architecture.

Before Maryland attempts to regulate healthcare AI comprehensively, it should first answer a more uncomfortable question:

How is artificial intelligence already being used inside Maryland’s own healthcare agencies? Are predictive algorithms already influencing Medicaid management, reimbursement review, audits, fraud detection or resource allocation?

The uncomfortable truth is that AI regulation has become politically attractive precisely because most people — including many policymakers — do not fully understand the underlying technology.

That creates the risk of symbolic legislation substituting for meaningful governance.

Maryland deserves credit for recognizing that artificial intelligence will fundamentally reshape healthcare. The state is right to ask difficult questions about bias, transparency, patient safety, accountability and insurer overreach.

But if Maryland wants to position itself as a national leader on healthcare AI governance, it must first confront a larger strategic question: Does the state actually possess the infrastructure, workforce and technical expertise necessary to lead?

Maryland is not a small player in healthcare or biotechnology. The state has one of the highest concentrations of physicians, academic medical centers, federal health agencies, biotech firms and research institutions in the country. Institutions connected to the National Institutes of Health, Johns Hopkins, the University System of Maryland and the broader I-270 biotechnology corridor have helped make Maryland a major center for genomics, biotechnology, life sciences and biomedical innovation.

The state is also deeply intertwined with federal healthcare infrastructure through agencies such as the NIH, FDA, CMS and other federal health and research entities located in or around Maryland.

In theory, Maryland should be uniquely positioned to lead the national conversation on healthcare AI.

That is where the conversation becomes far more complicated.

Maryland’s healthcare system, like much of the country, continues to struggle with physician shortages, nursing shortages, fragmented interoperability, aging public-sector IT infrastructure, delayed administrative systems, reimbursement pressures, cybersecurity vulnerabilities and mounting financial strain across hospitals and healthcare networks. Even large healthcare systems continue to experience enormous operational inefficiencies despite years of digital transformation initiatives that were supposed to modernize healthcare delivery.

At the same time, the state itself faces persistent fiscal pressures and structural budget deficits that raise an unavoidable question that lawmakers rarely address publicly: Does Maryland realistically possess the financial resources and technical expertise required to regulate highly sophisticated AI systems in healthcare at the level policymakers are promising?

Meaningful oversight of artificial intelligence is not a matter of creating a task force, issuing guidance documents or holding legislative hearings.

It requires a workforce composed of individuals who understand machine learning architecture, algorithmic auditing, health informatics, software engineering, predictive analytics, cybersecurity, clinical operations, statistics, data governance and the behavior of large language models operating inside complex healthcare environments.

AI adoption across healthcare is accelerating rapidly and will continue to do so because the economic, administrative and clinical pressures driving adoption are simply too significant to ignore.

It will require physicians, computer scientists, behavioral experts, data engineers, ethicists, insurers, regulators and frontline clinicians working together to build regulatory frameworks grounded in technological reality rather than political abstraction.

Otherwise, Maryland risks regulating only the visible surface of healthcare AI while ignoring the far more consequential infrastructure underneath it.

The elephant in the room is that we need leaders making AI policy who actually understand the reality of AI in clinical practice, not on paper.

Sreedhar Potarazu is an ophthalmologist and former health care executive.