AI_101_A_Physicians

AI 101: A Physician’s Guide -- A helpful introduction to artificial intelligence in healthcare

Prepared by the HCMS Artificial Intelligence Committee

DISCLAIMER: Please note that the content provided herein is informational only and should NOT, in any way, be considered legal, professional, business, practice, or other advice. Nor should any information found herein be considered a referral or promotion for any specific action or inaction. Consult your own practice adviser or attorney before taking any action or inaction based on this information.

Introduction

Artificial intelligence is already appearing throughout medicine — often quietly — through documentation tools, patient messaging systems, coding support, imaging software, scheduling systems, and payer workflows. That reality can create two equal and opposite problems: overconfidence and unnecessary fear. This guide is intended to give busy physicians a practical baseline: what AI is, what it is not, what basic Texas legal requirements apply, and what common risks deserve attention. The main point throughout is simple: AI may assist the physician, but it does not replace the physician’s judgment, professional responsibility, or relationship with the patient.

Part 1. What AI Is — and Is Not

Part 2. Where Physicians Are Encountering AI Today

Part 3. Texas Legal and Regulatory Framework

Part 4. Liability and Risk Considerations

Part 5. Some Do’s and Don’ts

Part 6. What Physicians Should Consider Next

Conclusion

References

At a Glance

Many physician-facing AI tools today are assistive, not autonomous.
AI may help reduce administrative burden but may also introduce liability and privacy risks.
Texas law now requires physician review and certain patient disclosures regarding AI use.
This guide is educational and designed to reduce uncertainty surrounding AI in medicine.

Part 1. What AI Is — and Is Not

What physicians should know first

Artificial intelligence is a broad term for computer systems that perform tasks associated with human intelligence, such as pattern recognition, prediction, language processing, and content generation. In healthcare, that can include everything from ambient documentation and inbox drafting to image analysis and decision support.

In medicine, “augmented” is often the word used

The American Medical Association uses the term augmented intelligence to emphasize that the most useful physician-facing tools are designed to support clinicians rather than replace them. That framing is helpful because most tools currently seen in practice are better understood as assistants, copilots, or decision-support tools—not independent clinicians.[1]

What AI is not

AI is not a guarantee of accuracy, objectivity, or current medical knowledge. Different tools are trained on different data, updated on different schedules, and built for different purposes. For example, a documentation scribe, a consumer chatbot, and a clinically validated diagnostic support tool should not be treated as interchangeable.

Common misconceptions

“AI” means the same thing in every setting. It does not. Administrative AI, generative AI, predictive models, and FDA-regulated software can raise very different clinical, legal, and operational questions.
If a tool sounds confident, it is probably right. Not necessarily. AI systems can produce inaccurate, outdated, incomplete, or fabricated content.
Using AI transfers responsibility to the vendor. It does not. In clinical practice, physician oversight remains essential, especially when AI output affects things like documentation or care.
If a tool is popular, it must be HIPAA-compliant or clinically validated. Popularity is not proof of compliance, security, or performance. Physicians and practices should verify those issues before use.

Key takeaway
For most physicians, AI should be viewed as a tool to support—not replace—clinical judgment. As AI continues to evolve, physicians should educate themselves on its capabilities and limitations and use trusted resources, such as HCMS, to help guide informed decisions about its use in practice.

Return to Table of Contents

Part 2. Where Physicians Are Encountering AI Today

What physicians may likely see first

For most physicians, the first wave of AI is not a robot diagnosing patients. It is software layered into work they already do: documenting visits, summarizing charts, drafting patient instructions, responding to inbox messages, assisting with coding, or helping flag findings for review. That matters because the practical AI question for most physicians is not whether AI is coming to medicine. It is where it will show up first in everyday workflow.[2][3]

Ambient documentation and note generation

One visible physician-facing category is ambient documentation: tools that listen to the patient encounter and generate a draft note, after-visit summary, or structured documentation for review. These tools are attractive because they aim to reduce clerical burden, but they still require physician review for accuracy, completeness, and context.[4][2]

Inbox, chart, and communication assistants

A second common category is generative assistance inside the EHR or patient portal. This includes chart summarization, draft responses to patient messages, discharge instructions, visit summaries, and translation support. In both independent practices and large systems, these tools are likely to expand because they can allow modest time savings without displacing physician judgment.[2][5]

Imaging and diagnostic support

In hospitals and multispecialty groups, physicians are also likely to encounter AI through radiology, cardiology, dermatology, pathology, waveform analysis, and other software that helps interpret images, signals, or patterns. Some of these tools are regulated as medical devices and are intended to help identify findings, prioritize worklists, or support diagnostic review—not eliminate the physician’s role.[6][7][8]

What this may look like in different practice settings

In an independent practice, the common entry point is often a lower-risk, time-saving tool: ambient documentation, chart summarization, discharge-instruction drafting, message drafting, or translation support. These can be easier to pilot because the physician can compare the AI draft against the underlying encounter, chart, or final message before it goes out.[4][2] In a large health system, AI may also appear through enterprise workflows the individual physician did not select personally—for example, EHR-embedded drafting tools, imaging support, centralized message-routing assistance, or operational tools that influence what reaches the physician’s inbox and when. That means some physicians will encounter AI even if they never go shopping for a stand-alone product.[3][5][8]

A practical way to think about products worth exploring

For most physicians, an initial question to consider is not “Which AI product is best?” but “Which part of my day is most repetitive, most time-consuming, and still appropriate for human review?” In 2026, the most widely available categories to explore are documentation assistants, chart and message drafting tools, translation support, coding and administrative assistants, and—depending on specialty and setting—imaging or diagnostic support tools. The key is to start with a use case where the physician can easily verify the output before it enters the chart or affects care.[2][3][8]
Return to Table of Contents

Part 3. Texas Legal and Regulatory Framework

Note: This section summarizes enacted Texas law as of May 8, 2026. Please note that the content provided herein is informational only, and should NOT, in any way, be considered legal, professional, business, practice, or other advice. Nor should any information found herein be considered a referral or promotion for any specific action or inaction. Consult your own practice adviser or attorney before taking any action or inaction based on this information.

SB 1188: electronic health records and AI

Effective Sept. 1, 2025, Texas Health & Safety Code Chapter 183 allows a health care practitioner to use AI for diagnostic purposes if the use is within the scope of the practitioner’s license, is not otherwise prohibited by state or federal law, and the practitioner reviews all records created with AI in a manner consistent with Texas Medical Board medical-record standards. The statute also requires disclosure to patients when AI is used for those diagnostic purposes.[9]

HB 149 (TRAIGA): disclosure in health care settings

Effective Jan. 1, 2026, the Texas Responsible Artificial Intelligence Governance Act requires disclosure when an AI system is used in relation to health care service or treatment. Under the statute, the disclosure must be provided to the patient or personal representative no later than the date the service or treatment is first provided, except in an emergency, when disclosure may be made as soon as reasonably possible. The law also says the disclosure must be clear, conspicuous, and written in plain language; a hyperlink may be used.[10]

What this means in practice

For many practices, one approach is a written disclosure that is easy to see and easy to understand, paired with a workflow-specific explanation when AI meaningfully affects documentation or care delivery. Because the Texas statutes overlap, practices should align their disclosure language and workflows with counsel or compliance personnel.

Physician responsibility remains unchanged

These statutes do not make AI the decision-maker. To the contrary, the Texas framework expressly requires physician review in certain settings and preserves the physician’s role in evaluating whether an AI-generated suggestion, note, or recommendation is appropriate for the patient in front of them.[9][10]

A practical disclosure standard
If a patient might inquire whether AI was involved in their care, the notice should be easy to find, easy to understand, and not buried in general paperwork. A brief explanation is usually more reassuring.

Return to Table of Contents

Part 4. Liability and Risk Considerations

Documentation risks

AI documentation tools can create problems when they do things like inventing facts, omitting relevant details, confusing speakers, misreading context, or failing during part of an encounter. A polished note can still be an inaccurate note. AI is also capable of “hallucinations”, which are inaccurate outputs, and recent work in clinical document generation has focused specifically on the need to verify factual accuracy before these outputs are used in patient care.[11]

Bias and accuracy gaps

Performance may vary across patient populations, accents, dialects, and care settings. Speech-recognition systems and other models can perform less reliably for some groups than others, which means a note or recommendation that looks routine on the screen may still contain unequal error risk.[12]

Over-reliance on AI output

The more familiar a tool becomes, the easier it is to review it less carefully. That creates automation bias: the tendency to trust or accept automated suggestions too readily, especially under time pressure or workload stress. In medicine, that can turn into things like signing a note too quickly, accepting a suggestion without checking the chart, or allowing the technology to narrow clinical thinking.[13]

Vendor limitations

Vendors may market a tool as efficient, intuitive, or “clinically informed,” but those claims do not answer the most important physician questions: What data trained the model? How was it validated? On whom was it validated? Can the vendor explain how the system reaches its output? What happens to protected health information (PHI) entered into the system? If PHI will flow through the tool, does the vendor offer terms appropriate for HIPAA-regulated use and, when required, a business associate agreement? What is the downtime plan? What happens if the company is acquired, changes terms, or stops supporting the product?[15][16]

A practical risk mindset

Consider starting with lower-risk uses, defining who reviews every output, monitoring for recurring errors, and making sure your practice knows how to pause or stop using a tool if the output becomes unreliable. Good governance is often less about saying yes or no to AI and more about knowing when human review must slow the process down.[14]

Before relying on an AI tool, consider the following:

What exactly is this tool supposed to do—and what is it not supposed to do?
Does it use PHI, and if so, what contractual and security protections are in place?
Who reviews the output every time?
What evidence supports its performance in my specialty, workflow, and patient population?
Can my practice function safely if the tool is unavailable tomorrow?

Return to Table of Contents

Part 5. Some Do’s and Don’ts

For busy clinicians, some of the very safeguards are usually the simplest ones: review the output, know the tool’s limits, protect patient information, and keep the physician in charge.

DO	DON’T
Review AI-generated content before it enters the medical record.	Do not assume polished output is accurate output.
Use AI to support--not replace--clinical judgment.	Do not paste identifiable patient information into open consumer platforms that are not approved by your practice.
Understand what data source, model type, and intended use sit behind the tool.	Do not use a general-purpose tool to aid in clinical decision-making unless it has been appropriately validated for that use.
Confirm privacy, security, and contracting requirements before entering PHI.	Do not treat vendor marketing as the same thing as evidence.
Disclose AI use to patients when required and consider brief, plain-language explanations when appropriate.	Do not skip review simply because the tool has been reliable in the past.
Monitor for recurring errors, especially in complex visits or with speech-recognition tools.	Do not ignore fairness, population fit, or downtime planning.

One practical bottom line If you would not feel comfortable explaining the tool’s role to a patient, a colleague, or a reviewer after the fact, the workflow likely needs more human oversight, better documentation, or both.

Return to Table of Contents

Part 6. What Physicians Should Consider Next

Insurer use of AI

Some of the most consequential AI developments for physicians may come from payers rather than from clinical tools.[17] In 2025, the NAIC reported that 84% of surveyed health insurers used Artificial Intelligence/Machine Learning in some part of operations, but the underlying survey results show more variation when one looks at specific functions such as prior authorization. In other words, the payer AI story is real, but not every insurer is using AI in the same way or at the same point in the workflow.[18][19]

Why this matters in practice

For physicians, the concern is not simply whether an insurer uses AI. Some of the more important questions are how the tool is used, what human review occurs, and how easy it is to challenge a questionable decision. KFF reported that Medicare Advantage insurers made nearly 53 million prior authorization determinations in 2024, denied 7.7% of requests, saw only 11.5% of denials appealed, and overturned 80.7% of appealed denials at the first level.[20] That pattern does not prove AI caused a particular denial. It does show why physicians should pay attention to transparency, documentation, and appeal pathways whenever an automated or semi-automated process is involved.

Traditional Medicare is now part of the conversation

The CMS WISeR model, which began in 2026 in selected Original Medicare regions, including Texas, is worth watching closely.[21] CMS describes WISeR as a technology-enhanced prior authorization and pre-payment review model for a limited set of services vulnerable to waste, fraud, abuse, or low-value use. CMS also states that the model does not change Medicare coverage or payment policy, excludes emergency and inpatient-only services, and requires appropriately licensed clinicians to determine recommendations for non-payment.[21] For physicians, the practical question is whether models like this reduce inappropriate care and administrative burden—or simply move more utilization review upstream.

Litigation and state oversight are accelerating

Courts and regulators are beginning to look more closely at insurer algorithms, governance, and human oversight. This signals that insurer AI governance is becoming a legitimate subject of legal scrutiny. At the state level, California’s SB 1120 prohibits AI or similar tools from denying, delaying, or modifying care based on medical necessity and by requiring those determinations to be made by a licensed physician or other qualified licensed health care professional competent to evaluate the specific clinical issues involved.[23] Nationally, the NAIC is piloting an AI Systems Evaluation Tool in 2026 to help regulators assess insurer governance, transparency, and risk management.[24]

Federal regulatory developments

On the clinical side, FDA oversight continues to develop along a risk-based path. Some clinical decision support software may fall outside device regulation if it allows a health care professional to independently review the basis for the recommendation and not rely primarily on the software’s output.[6] By contrast, software that analyzes medical images, physiologic signals, or otherwise functions as a medical device remains subject to FDA oversight.[7][8] FDA has also issued guidance on predetermined change control plans for artificial intelligence-enabled device software functions, signaling a more structured lifecycle approach to updates and monitoring.[25] Separate from device oversight, FDA and the European Medicines Agency jointly published ten principles for good AI practice in drug development in 2026, underscoring expectations around human oversight, risk management, data quality, and documentation.[26]

Return to Table of Contents

Conclusion

For many physicians, the next important AI issue will not be whether their own practice adopts a new tool. It will be how AI is used around the practice—by payers, vendors, regulators, and business partners whose decisions can affect access, documentation, and workload. A practical response is to document medical necessity clearly, preserve records when a denial appears automated or poorly explained, and watch for new state or federal rules that require additional obligations for you.

Selected References

[1] American Medical Association. “Augmented Intelligence in Medicine.” https://www.ama-assn.org/practice-management/digital-health/augmented-intelligence-medicine

[2] American Medical Association. “2026 Physician Survey on Augmented Intelligence.” March 2026. https://www.ama-assn.org/system/files/physician-ai-sentiment-report.pdf

[3] Assistant Secretary for Technology Policy/Office of the National Coordinator for Health Information Technology. “SAFER for EHR Resilience: High Priority Practices.” August 2024 revision. https://healthit.gov/wp-content/uploads/2025/06/SAFER-Guide-4.-High-Priorities-Final.pdf

[4] Tierney AA, et al. “Ambient Artificial Intelligence Scribes to Alleviate the Burden of Clinical Documentation.” NEJM Catalyst, 2024. https://catalyst.nejm.org/doi/pdf/10.1056/CAT.23.0404

[5] Assistant Secretary for Technology Policy/Office of the National Coordinator for Health Information Technology. “SAFER for EHR Resilience: Clinician Communication.” August 2024 revision. https://healthit.gov/wp-content/uploads/2025/06/SAFER-Guide-1.-Clinical-Communication-Final.pdf

[6] U.S. Food and Drug Administration. “Clinical Decision Support Software.” https://www.fda.gov/regulatory-information/search-fda-guidance-documents/clinical-decision-support-software

[7] U.S. Food and Drug Administration. “Artificial Intelligence in Software as a Medical Device.” https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-software-medical-device

[8] U.S. Food and Drug Administration. “Artificial Intelligence-Enabled Medical Devices.” https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices

[9] Texas Legislature. S.B. 1188 (89th Legislature, enrolled version), Health & Safety Code Chapter 183. https://capitol.texas.gov/tlodocs/89R/billtext/html/SB01188F.htm

[10] Texas Legislature. H.B. 149 (89th Legislature, enrolled version), Texas Responsible Artificial Intelligence Governance Act. https://capitol.texas.gov/tlodocs/89R/billtext/html/HB00149F.htm

[11] Chung P, et al. “Verifying Facts in Patient Care Documents Generated by Large Language Models.” NEJM AI, 2025. https://ai.nejm.org/doi/full/10.1056/AIdbp2500418

[12] Martin JL, Wright KE. “Bias in Automatic Speech Recognition: The Case of African American Language.” Applied Linguistics, 2023. https://academic.oup.com/applij/article/44/4/613/6901317

[13] Goddard K, et al. “Automation bias: a systematic review of frequency, effect mediators, and mitigators.” Journal of the American Medical Informatics Association, 2011. https://pmc.ncbi.nlm.nih.gov/articles/PMC3240751/

[14] National Institute of Standards and Technology. “AI Risk Management Framework.” https://www.nist.gov/itl/ai-risk-management-framework

[15] U.S. Department of Health and Human Services. “Business Associate Contracts.” https://www.hhs.gov/hipaa/for-professionals/covered-entities/sample-business-associate-agreement-provisions/index.html

[16] U.S. Department of Health and Human Services. “Guidance on HIPAA and Cloud Computing.” https://www.hhs.gov/hipaa/for-professionals/special-topics/health-information-technology/cloud-computing/index.html

[17] Mello MM, Adler-Milstein J, et al. “The AI Arms Race in Health Insurance Utilization Review.” Health Affairs, 2026. https://www.healthaffairs.org/doi/10.1377/hlthaff.2025.00897

[18] National Association of Insurance Commissioners. “NAIC Survey Reveals Majority of Health Insurers Embrace AI.” May 20, 2025. https://content.naic.org/article/naic-survey-reveals-majority-health-insurers-embrace-ai

[19] National Association of Insurance Commissioners. “Health Insurance Artificial Intelligence/Machine Learning Survey Results.” May 9, 2025. https://content.naic.org/sites/default/files/inline-files/Health%20Survey%20Report%20-%20FINAL%205.9.25.pdf

[20] KFF. “Medicare Advantage Insurers Made Nearly 53 Million Prior Authorization Determinations in 2024.” January 28, 2026. https://www.kff.org/medicare/medicare-advantage-insurers-made-nearly-53-million-prior-authorization-determinations-in-2024/

[21] Centers for Medicare & Medicaid Services. “WISeR Model Provider and Supplier Guide.” April 24, 2026. https://www.cms.gov/priorities/innovation/files/wiser-provider-supplier-guide.pdf

[22] Estate of Gene B. Lokken et al. v. UnitedHealth Group, Inc. et al., Order on Motion to Compel Discovery (D. Minn. Mar. 9, 2026). https://litigationtracker.law.georgetown.edu/wp-content/uploads/2023/11/Estate-of-Gene-B.-Lokken-et-al-v.-UnitedHealth-Group-Inc.-et-al_2026_3_9_ORDER-ON-MOTION-TO-COMPEL-DISCOVERY.pdf

[23] California Department of Insurance. “CDI Guidance SB 1120: Use of Artificial Intelligence, Algorithms and Other Software Tools in Utilization Management.” May 5, 2025. https://www.insurance.ca.gov/0250-insurers/0500-legal-info/0200-regulations/HealthGuidance/upload/SB-1120-1-Guidance-Use-of-Artificial-Intelligence-Algorithms-and-Other-Software-Tools-in-Utilization-Management.pdf

[24] National Association of Insurance Commissioners. “AI Systems Evaluation Tool Pilot: Pilot Project Background.” https://content.naic.org/sites/default/files/call_materials/Pilot%20Project%20Summary.pdf

[25] U.S. Food and Drug Administration. “Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions.” https://www.fda.gov/regulatory-information/search-fda-guidance-documents/marketing-submission-recommendations-predetermined-change-control-plan-artificial-intelligence

[26] U.S. Food and Drug Administration. “Guiding Principles of Good AI Practice in Drug Development.” January 14, 2026. https://www.fda.gov/about-fda/artificial-intelligence-drug-development/guiding-principles-good-ai-practice-drug-development