The uncomfortable question every independent physician should ask before adopting AI: When technology learns from your practice, who benefits from what it learns?

“The good physician treats the disease; the great physician treats the patient who has the disease.” — Sir William Osler

The Great Healthcare AI Debate: Who Owns the Intelligence?

Imagine this.

You spend 20 years building a successful medical practice.

You develop better workflows. You train your staff. You learn which insurance requirements create problems. You discover documentation patterns that prevent coding errors. You figure out how to get claims paid without spending half your life chasing insurance companies.

Then you hire AI.

It promises to reduce administrative work, improve billing accuracy, automate repetitive tasks, and give you back your evenings.

Wonderful.

But here is an uncomfortable question:

Are you buying intelligence, or are you contributing your hard-earned intelligence to someone else’s business?

Imagine your AI vendor becoming smarter from the operational knowledge generated by thousands of medical practices.

Now imagine that you have no clear understanding of what happens to your information, how long it is retained, whether it is used for model training, or who can access it.

Your competitor down the street might be using the same vendor.

Are you both benefiting from the same technology?

Or are you unknowingly helping build an ecosystem in which the vendor owns the intelligence and the physician merely rents access?

Let’s be clear. A cloud AI model does not automatically learn and redistribute your billing strategies every time you submit a prompt. Training, inference, retention, and data sharing are different technical processes.

But that distinction makes the question even more important.

If you do not understand what happens to your data, you do not fully understand the technology you purchased.

And that is a dangerous way to run a medical practice.

Welcome to the new healthcare debate.

The issue is no longer whether physicians should use AI.

The issue is whether physicians will remain in control of the intelligence that AI creates.

The Biggest Problem in Healthcare May Not Be What We Think

Physicians are drowning in administration.

They spend years learning medicine, only to discover that running a practice also requires expertise in insurance eligibility, prior authorization, documentation requirements, coding, claim submission, payer policies, and reimbursement.

Apparently, medical school forgot to include a degree in insurance-company diplomacy.

A typical encounter may involve:

• Patient registration.
• Insurance verification.
• Referral review.
• Authorization.
• Clinical documentation.
• Medical decision-making.
• Coding.
• Charge capture.
• Claim submission.
• Payer adjudication.
• Payment reconciliation.

One missing piece of information can travel through this entire process before anyone notices.

And then everyone gets involved.

The front desk investigates.

The billing department investigates.

The physician receives a query.

The office manager investigates.

Eventually, someone discovers that the problem began with information that was incomplete three weeks earlier.

Congratulations. You have successfully created a team-building exercise nobody wanted.

But here is the bigger problem.

We have become very good at repairing problems that should never have happened.

A coding query may originate from incomplete documentation.

A rejection may originate from incorrect patient or provider information.

A denial may involve eligibility, coverage, authorization, coding, or medical necessity.

A payment delay may originate much earlier in the workflow.

These are different problems and require different corrective actions.

A clearinghouse rejection is not the same as a payer denial. One generally occurs before payer adjudication; the other follows the payer’s adjudication process.

Yet practices often treat every unpaid claim as another item on the follow-up list.

That is like repeatedly mopping the floor while refusing to repair the leaking pipe.

The best billing problem is the one that never happens.

This is the difference between reactive billing and upstream billing intelligence.

Cloud AI Versus Private AI: A False Choice?

The industry is debating two approaches.

One side says healthcare needs cloud AI, connected EHRs, interoperability, and shared intelligence.

The other says practices need private AI, stronger infrastructure isolation, and control over proprietary information.

Both have a point.

Both can also be wrong.

The cloud argument

Healthcare information is fragmented.

One system contains the clinical record.

Another handles scheduling.

Another manages billing.

Another handles laboratory results.

Another handles insurance information.

Standards such as HL7 and FHIR help these systems exchange information.

AI can help interpret unstructured documentation, organize information, and support workflows across different platforms.

That is valuable.

But let’s challenge one assumption.

Connecting more systems does not automatically create better healthcare.

If the underlying information is inaccurate, inconsistent, or incomplete, connecting everything simply allows bad information to travel faster.

Garbage in. Faster garbage out.

AI is not a magic translator that can repair every broken interface or determine clinical truth from incomplete documentation.

Interoperability requires standards, integration, identity matching, terminology mapping, permissions, and validation.

The private AI argument

Private AI offers another approach.

A practice may choose dedicated infrastructure, a private cloud deployment, or a locally hosted model.

This can improve control over data access and processing.

It may also provide predictable pricing in certain arrangements.

But private AI is not automatically secure.

A server in a locked closet is still vulnerable if nobody patches it, monitors it, backs it up, or knows how to recover it.

A small medical practice does not necessarily need to become its own technology company.

Owning a server does not make you a cybersecurity expert.

The answer is not to choose between connection and isolation.

The answer is to separate interoperability from unrestricted data use.

A medical practice should be able to exchange information with an authorized hospital without automatically allowing a commercial AI vendor to reuse its proprietary operational information.

Connect the information. Protect the intelligence.

That is the architecture worth discussing.

The Statistics: Physicians Are Already Moving Toward AI

The AI revolution is not waiting for medical practices to catch up.

The American Medical Association’s 2026 Physician Survey on Augmented Intelligence reported:

• 81% of physicians use AI professionally, compared with 38% in 2023.
• 86% identified data privacy assurances as important to AI adoption.
• 88% emphasized the importance of robust safety and efficacy validation.
• 85% wanted physicians involved in decisions about AI adoption.
• 70% saw opportunities for AI to automate tasks contributing to burnout.
• 40% reported being equally excited and concerned about AI.

The numbers reveal something interesting.

Physicians are not necessarily afraid of AI.

They are afraid of adopting something they cannot adequately evaluate, control, or trust.

And frankly, they should ask questions.

A software demonstration is not clinical validation.

A HIPAA statement is not a complete security assessment.

A productivity claim is not proof of financial benefit.

Physicians do not need more AI promises. They need fewer administrative problems.

Three Experts. Three Lessons. One Important Message.

Dr. John Whyte — American Medical Association

Lesson: AI should support physicians, not displace their judgment.

Dr. Whyte and the AMA have emphasized responsible AI adoption, physician involvement, and the importance of trust.

The 2026 physician survey reinforces the importance of privacy, safety, and professional oversight.

My interpretation:

Do not purchase AI because it promises to replace human work.

Purchase it because it can improve the conditions under which humans perform that work.

A faster wrong answer is still wrong.

Practical advice: Establish which decisions AI may recommend, which actions it may perform, and which decisions require physician or staff approval.

Dr. Jonathan Perlin — Coalition for Health AI

Lesson: Governance must be part of the product, not an afterthought.

The Coalition for Health AI has developed governance resources addressing organizational oversight, risk assessment, responsible data management, vendor management, and education.

For a small practice, this does not mean hiring an army of compliance officers.

It means knowing who is responsible for AI, what the system can access, and how problems are investigated.

Practical advice: Document AI ownership, permissions, intended use, oversight, and escalation procedures.

If nobody owns the decision, nobody owns the risk.

Ada Tsoi, PhD — UNC Health

Lesson: Evaluate the application, not just the technology.

CHAI’s AI assessment work emphasizes evaluating specific use cases and the risks associated with actual deployment.

A system that summarizes administrative information is not necessarily equivalent to one that accesses patient records or influences clinical decisions.

The product name alone tells you very little about its real-world risk.

Practical advice: Test the actual workflow, permissions, and outputs before granting broad access.

The risk is not simply what AI can do. It is what you allow it to do.

Five Myths About AI in Medical Billing

MYTH 1: HIPAA compliance means AI is automatically safe.

Reality: HIPAA obligations, business associate agreements, security safeguards, retention, and permitted data use must be evaluated in context.

A BAA is important when required. It is not a magic shield.

MYTH 2: Every cloud AI model learns from every customer.

Reality: No. Vendors have different policies. Some restrict training on customer data. Others may have different retention or product-improvement arrangements.

Ask for written terms.

MYTH 3: Private AI is always better.

Reality: Security depends on implementation, monitoring, maintenance, and recovery.

A poorly managed private server can be less secure than a professionally managed cloud environment.

MYTH 4: Automating billing eliminates billing problems.

Reality: Automation can accelerate bad processes.

Validate information before allowing AI to propagate it.

MYTH 5: Denial management is the ultimate solution.

Reality: Recovering denied claims is important. Preventing recurring upstream errors is equally important.

Stop celebrating how efficiently your practice fixes the same mistake. Start asking why it keeps happening.

Recent News: The AI Agent Is No Longer Just a Chatbot

One of the most important developments in healthcare technology is the transition from conversational AI to autonomous AI agents.

A chatbot responds.

An agent may access information, create tasks, send messages, update workflows, and execute authorized actions.

That changes the risk equation.

In November 2026, Reuters reported on Australia’s plans for stronger AI incident reporting and oversight following an incident involving an AI agent accessing a government Medicare medical statistics portal without authorization.

The incident should not be confused with evidence that patient records were stolen.

But it illustrates a larger issue.

An AI agent can interact with systems beyond the immediate conversation.

That requires careful permissions and monitoring.

The practical lesson for medical practices:

Do not give an AI agent the keys to the building because you asked it to open one door.

CHAI’s October 2026 AI Intake Playbook also offers a useful development. Its assessment framework includes standardized controls and evidence requirements to help organizations evaluate AI vendors more consistently.

Meanwhile, healthcare cybersecurity guidance increasingly emphasizes third-party risk, incident response, and business continuity.

The message is becoming difficult to ignore.

AI governance is no longer just an enterprise IT issue.

It is a business survival issue for independent medicine.

The Six Most Expensive Pitfalls to Avoid

Before signing another software agreement, consider these questions.

1. Who owns the information?

Clarify ownership, permitted uses, generated records, and portability.

2. Is our data used for model training?

Ask whether prompts, outputs, clinical information, or workflow data are retained or used for training or product improvement.

3. Who has access?

Identify vendor personnel, subcontractors, and external integrations.

4. Can we audit the system?

Demand meaningful logs showing access, recommendations, and actions.

5. What happens when the vendor goes down?

Can your staff continue essential operations?

6. Can we leave?

Can you retrieve your information, revoke access, and transition without losing essential records?

One more thought:

A vendor’s real value is not measured only by what happens when everything works. It is measured by what happens when something fails.

Legal and Ethical Implications: Technology Does Not Transfer Responsibility

AI does not automatically transfer a medical practice’s legal obligations to its software vendor.

Physicians should consider:

• HIPAA privacy and security requirements.
• Business associate agreements where applicable.
• California medical information privacy requirements.
• Applicable state and federal consumer privacy laws.
• Data retention and deletion.
• Third-party subcontractors.
• Documentation integrity.
• Coding and billing accuracy.
• Human oversight.
• Incident notification and recovery.

Clinical documentation must accurately represent the encounter.

AI should not invent findings, manufacture medical necessity, or encourage unsupported coding.

Ethically, patient confidentiality, fairness, transparency, and physician autonomy remain central.

The algorithm may generate the recommendation. The practice must understand who is accountable for the action.

A Practical Ten-Step AI Adoption Strategy

A three-physician practice does not need a million-dollar IT department.

It needs discipline.

Step 1: Find the pain. Identify the recurring administrative problem that consumes the most time.

Step 2: Measure it. Establish baseline error rates, staff hours, and financial consequences.

Step 3: Map the workflow. Find where information originates, changes, becomes incomplete, or disappears.

Step 4: Classify the data. Separate protected health information, operational data, and proprietary business intelligence.

Step 5: Evaluate the vendor. Review security, model training, retention, subcontractors, and contractual terms.

Step 6: Start small. Pilot one workflow using synthetic or appropriately de-identified data where possible.

Step 7: Keep humans involved. Define which recommendations require approval.

Step 8: Monitor performance. Measure accuracy, rework, and unintended consequences.

Step 9: Prepare for failure. Establish access revocation, downtime procedures, backups, and recovery plans.

Step 10: Expand only when evidence supports it. Do not automate ten workflows before proving one works.

What Should Physicians Actually Measure?

Forget the number of AI features in the sales presentation.

Measure operational outcomes.

Useful metrics include:

• First-pass claim acceptance.
• Clearinghouse rejection rate.
• Payer denial rate.
• Clean claim rate.
• Days in accounts receivable.
• Eligibility error frequency.
• Authorization turnaround time.
• Documentation query rate.
• Staff rework hours.
• Manual data-entry volume.
• AI recommendation correction rate.
• Vendor downtime.
• Security incidents.
• Recovery time.

Separate technical rejections from payer denials.

Separate documentation problems from authorization problems.

Separate preventable errors from legitimate payer disputes.

If you cannot measure the problem, you cannot prove the technology solved it.

The OnnX Perspective: The Best Billing Problem Is the One That Never Happens

This is where the debate becomes personal for me as a physician-entrepreneur.

I have spent considerable time thinking about why medical practices repeatedly encounter the same administrative and billing problems.

And I keep returning to one uncomfortable observation.

We have built an industry around fixing problems that often begin before the billing department ever sees them.

What if we changed the starting point?

What if the objective was not simply to process claims faster, but to improve the information that eventually becomes a claim?

That is the thinking behind OnnX 2.0.

OnnX is an upstream intelligence platform for healthcare, built by physicians and medical billers.

Its philosophy is simple:

Reduce downstream administrative and billing noise by improving the quality, structure, and continuity of information upstream.

We are interested in the relationship between:

• Patient registration.
• Insurance eligibility.
• Scheduling.
• Authorization.
• Clinical encounters.
• Documentation.
• Coding.
• Charge capture.
• Claims.
• Payer requirements.
• Payment outcomes.

These are not isolated transactions.

They are connected pieces of one operational process.

The OnnX approach is built around six principles:

Capture → Structure → Preserve → Propagate → Act → Learn

Capture relevant information.

Structure it.

Preserve its context.

Propagate validated information to authorized workflows.

Act when a problem is identified.

Learn from outcomes.

The objective is not to become another traditional RCM service provider.

It is to develop upstream intelligence that helps practices recognize recurring problems before they become downstream billing work.

Consider the difference.

Traditional question:

“Why wasn’t this claim paid?”

Upstream question:

“Why did the information that created this claim become incomplete or inconsistent in the first place?”

One question looks backward.

The other creates an opportunity to improve the next encounter.

A rescued claim is a recovery. A claim that never needed rescuing is progress.

That is the philosophy behind OnnX 2.0.

And it is why we believe independent practices deserve technology that helps them retain control of their operational intelligence.

The Future: From Administrative Automation to Intelligent Medicine

The future of healthcare will not be determined by the biggest AI model alone.

It will depend on the quality of information, the architecture connecting systems, and the safeguards governing decisions.

AI agents will increasingly coordinate administrative workflows.

EHRs will become more connected.

Clinical and operational data will become more structured.

Revenue-cycle systems will move toward earlier detection of problems.

But there is a danger.

We could spend billions automating broken processes and call it innovation.

We could create faster documentation queries, faster denials, faster appeals, and faster administrative chaos.

Or we could rethink the process itself.

The future is not a faster billing department. It is a medical practice that needs less billing repair.

Physicians should not have to choose between innovation and independence.

They should not have to sacrifice privacy to obtain interoperability.

And they should not have to become cybersecurity engineers just to purchase useful software.

The next generation of healthcare technology must be practical, secure, accountable, and financially sustainable.

That is a much higher standard than simply adding an AI button to another application.

Frequently Asked Questions

1. Is cloud AI safe for medical practices?

It can be, depending on the vendor, configuration, contractual terms, access controls, and intended use. Evaluate the actual deployment rather than assuming all cloud platforms are equally secure.

2. Does HIPAA prevent AI vendors from using patient information?

HIPAA imposes restrictions on covered entities and business associates. Permitted uses depend on the applicable legal framework and agreements. A BAA does not automatically authorize every form of data use.

3. Does private AI eliminate data leakage?

No. It can provide greater infrastructure control, but security still depends on access management, maintenance, monitoring, and incident response.

4. Does a cloud AI model learn from my billing data?

Not necessarily. Training policies differ. Ask whether information is retained, used for training, used for product improvement, or shared with subcontractors.

5. Should a small medical practice install its own AI server?

Not automatically. Compare private hosting, managed private cloud, and appropriately governed commercial cloud services based on actual security, cost, and operational requirements.

6. Can AI eliminate medical billing denials?

No. It may help prevent certain avoidable errors, but coverage disputes, medical necessity determinations, payer policies, and other legitimate issues will remain.

7. What is upstream billing intelligence?

It is the use of structured information and intelligent workflows to identify and prevent administrative problems before they become coding, claim, or payment problems.

8. How should a practice start?

Choose one recurring problem, establish a baseline, pilot a controlled solution, measure results, and expand only when evidence supports it.

Final Thoughts: Stop Renting Intelligence Without Understanding What You Own

There is a bigger issue behind the healthcare AI debate.

For decades, independent physicians have depended on external systems to manage increasingly complicated administrative processes.

Now AI promises to make those systems smarter.

That is exciting.

But smarter systems do not automatically create more independent practices.

The question is whether physicians will understand and control the information, decisions, and workflows that define their businesses.

We should welcome innovation.

We should demand interoperability.

We should protect patient information.

We should challenge vendors.

And we should stop confusing automation with progress.

The greatest competitive advantage in medicine may not be the AI you purchase. It may be the intelligence you preserve.

Get Involved: Help Shape the Future of Independent Medicine

I have three questions for physicians, clinic owners, administrators, and medical billing professionals.

1. Provoke: If your AI vendor became smarter from your practice’s operational experience, would you know exactly what it learned, what it retained, and who benefits?

2. Comment: What concerns you most: AI costs, data privacy, vendor dependency, or the administrative problems that AI is supposed to solve?

3. Share: If you believe independent physicians deserve better technology and greater control, repost this article and bring another physician into the conversation.

Three things worth remembering:

Don’t automate a broken process. Fix the process first.

Don’t confuse access to intelligence with ownership of intelligence.

Don’t let the future of independent medicine be designed without independent physicians.

Let’s move from reactive billing to intelligent prevention.

Let’s build technology around the realities of medical practice.

Let’s make AI work for physicians, not the other way around.

About the Author

Dr. Daniel Cham is a physician, entrepreneur, and medical technology consultant with experience in clinical practice, healthcare management, and medical billing.

As founder of OnnX 2.0, he explores how AI, structured information, and upstream billing intelligence can help independent physician practices reduce administrative friction and strengthen operational sustainability.

His work focuses on the intersection of clinical care, technology, and the business of medicine.

Disclaimer / Note

This article is provided for general educational and informational purposes. It does not constitute legal, medical, cybersecurity, or financial advice. Healthcare organizations should consult qualified professionals regarding their particular regulatory, contractual, clinical, and operational circumstances.

References

1. The Patient Got Better. The Claim Got Denied. And Now AI Wants to Join the Argument.
   Explores healthcare AI, administrative inefficiency, accountability, and the importance of addressing problems before they become billing disputes.
2. Revolutionizing Medical Billing: How AI and Automation Are Enhancing Accuracy, Efficiency, and Reducing Denials
   Discusses AI vendor selection, data ownership, audit rights, contract terms, model transparency, and implementation safeguards.
3. Healthcare Doesn’t Have an AI Problem. It Has an Information Problem. And Your Billing Department Is Paying for It.
   Examines fragmented healthcare information, interoperability, data quality, and why better information must come before better automation.

4. Coalition for Health AI (CHAI): AI Intake Playbook and Common Intake Controls
   A healthcare-focused framework for evaluating AI vendors, assessing deployment risks, reviewing supporting evidence, and strengthening organizational AI governance.
5. Reuters: Australia to Mandate AI Breach Reporting and Create Regulator After OpenAI Incident
   A real-world example of why AI-agent permissions, cybersecurity controls, monitoring, and incident reporting matter. The reported incident involved unauthorized access to a Medicare medical statistics portal; it should not be described as a confirmed exposure of patient records.
6. BakerHostetler: Digital Health and AI—Key Regulatory, Reimbursement and Policy Developments
   Reviews evolving healthcare AI governance, vendor diligence, contracting, reimbursement, privacy, security, and accountability considerations.

Continue the Conversation

Medicine, technology, and business operations are changing together. Explore more perspectives, practical strategies, and discussions about the future of independent healthcare.

Connect: LinkedIn | Website | YouTube| Spotify

Knowledge is not valuable simply because we collect more of it. It becomes valuable when we use it to make better decisions. Keep questioning, keep learning, and help shape the future of medicine.

Start your journey toward better healthcare intelligence today.

PS: Check the Featured section of my LinkedIn profile for a free resource. No signup required.

If this perspective resonates, consider reposting it to help more physicians and clinic owners rethink the relationship between AI, medical billing, and practice independence.

#OnnX #OnnX2 #HealthcareAI #MedicalBilling #UpstreamBillingIntelligence #RevenueCycleManagement #PhysicianEntrepreneur #IndependentMedicine #MedicalPracticeManagement #HealthcareInnovation #ArtificialIntelligence #HealthcareCybersecurity #DataPrivacy #HIPAA #EHR #FHIR #ClinicalDocumentation #ValueBasedCare #PhysicianLeadership #HealthIT