AI in Clinical Diagnosis: Who Is Liable When the Algorithm Gets It Wrong?

As AI-assisted diagnostic tools move from pilot projects into everyday use across Indian hospitals, a question that used to feel theoretical is now genuinely practical: if an AI tool contributes to a missed or wrong diagnosis, who is legally responsible — the doctor who used it, the hospital that deployed it, or the company that built it? India does not yet have a dedicated AI liability law, which means the answer currently depends on stitching together several existing legal frameworks. This article walks through how that actually works today.

The Short Answer: The Doctor Remains the Anchor of Liability

Across every current legal framework that could apply, one principle holds consistently: if a doctor uses an AI tool, they remain responsible for exercising reasonable care, judgement, and skill — the standard of care applies exactly as it would without the tool involved. India’s official position on the clinical decision support systems now being deployed nationally is explicit on this point: these tools are designed to assist clinicians, not override their judgement, and the treating doctor retains final authority over diagnosis and treatment. This design choice has direct legal significance — because the doctor remains the final decision-maker, the doctor also remains the primary point of legal accountability in most scenarios.

How Existing Negligence Law Applies

The medical negligence framework covered in more depth elsewhere in this series — the “reasonable competence, not perfection” standard from Jacob Mathew v. State of Punjab, and the consumer protection framework from Indian Medical Association v. V.P. Shantha — applies without modification to AI-assisted care. A doctor who follows an AI recommendation without applying independent clinical judgement, or who fails to catch an error the AI tool made, can be found negligent under the existing standard: did the doctor exercise the care and skill a reasonably competent practitioner would have, given all the information available, including the AI’s output? An AI suggestion does not lower that bar, and courts are likely to treat blind over-reliance on an AI recommendation, without appropriate clinical correlation, as a failure of the doctor’s own independent judgement rather than an excuse.

Three Categories of AI Tools, Three Different Liability Pictures

• Assistive tools (most current Indian deployments): systems that flag patterns, suggest differentials, or provide evidence-based recommendations — like the national Clinical Decision Support System covered in a companion article — while leaving the actual diagnostic and treatment decision to the doctor. Here, liability sits overwhelmingly with the doctor, since the tool is explicitly a decision-support aid, not a decision-maker.
• Tools that provide reasoning or confidence scores alongside their output: these create a more genuinely shared responsibility picture between the doctor, the hospital that deployed the tool, and potentially the developer, since the system is doing more independent analytical work, though the doctor’s obligation to apply clinical correlation before acting remains.
• Fully autonomous diagnostic systems operating without clinician supervision: these remain rare and are generally unacceptable under current Indian medical practice standards — the Telemedicine Practice Guidelines, 2020, covered elsewhere in this series, explicitly do not permit AI platforms to independently conduct consultations, diagnose conditions, or prescribe medicines. Where such a system is used regardless, responsibility would likely shift more heavily toward the manufacturer and the institution that deployed it.

Where Product Liability and Consumer Protection Law Come In

Beyond medical negligence specifically, India’s Consumer Protection Act, 2019, includes product liability provisions (under Chapter VI) that can hold manufacturers, service providers, and sellers liable for harm caused by a defective product — and there is a reasonable legal argument that AI diagnostic software, when viewed as a product, could fall within this framework if it is shown to be defective in its design, data, or performance. Separately, a wrong diagnosis attributable to a flawed AI tool could be framed as a “deficiency in service” under consumer protection law, and the Central Consumer Protection Authority has the power to investigate and penalise misleading claims made about an AI tool’s capabilities. This is a developing area of Indian consumer law rather than a settled body of precedent specific to healthcare AI.

The “Black Box” Problem: Why This Is Harder Than It Sounds

A recurring, genuine legal complication is the “black box” nature of many AI systems — even the doctor or developer using the tool may not fully understand the specific reasoning behind a given output. This creates real difficulty in establishing causality: if an AI tool contributed to a missed diagnosis, was it because the training data was biased or unrepresentative, because the algorithm itself had a flaw, because the doctor failed to apply appropriate clinical correlation, or some combination of all three? Indian courts examining such a case would likely need to rely on expert testimony to unpack this in a way that traditional medical negligence cases, which turn on more straightforwardly observable clinical decisions, typically do not require.

What the ICMR’s Ethical Guidelines Say

The Indian Council of Medical Research issued Ethical Guidelines for the Application of Artificial Intelligence in Biomedical Research and Healthcare in 2023, which — while not itself a binding statute — sets out expectations that responsible institutions are increasingly treating as a practical governance framework. These guidelines call for model risk tiering, prospective validation before deployment, and post-deployment auditing of AI tools used in healthcare, along with oversight and ethical review of diagnostic AI systems and regular audits to check for bias and inaccuracy. A hospital or clinic deploying an AI diagnostic tool without any of this kind of governance in place would likely be viewed less favourably in any subsequent negligence or deficiency-in-service inquiry, even though the ICMR guidelines themselves are not directly enforceable as law.

Criminal Liability: A Narrower, Harder-to-Establish Category

In the most serious cases — where AI-assisted negligence causes a patient’s death — Section 106 of the Bharatiya Nyaya Sanhita, 2023 (the successor to IPC Section 304A) could theoretically apply, covering death caused by a rash or negligent act not amounting to culpable homicide. But where the negligence arises from an AI system’s malfunction or flawed data analysis rather than a clear-cut clinical error by the doctor, establishing the gross negligence threshold criminal liability requires becomes legally complicated — this is a genuinely unsettled area, and criminal liability specifically for AI-attributable errors remains rare and difficult to establish under current law.

How to Practically Protect Yourself as a Doctor Using AI Tools

1. Treat every AI output as a decision-support input, not a final answer — document that you applied independent clinical correlation before acting, particularly when the AI’s suggestion informed a consequential decision.
2. Understand, at least at a basic level, what the AI tool you’re using is actually designed to do, its known limitations, and whether it has undergone the kind of validation the ICMR guidelines describe — don’t treat it as an unquestionable black box.
3. Document disagreements with an AI recommendation, and your clinical reasoning for overriding it, just as thoroughly as you would document agreement with it — this protects you either way.
4. If your hospital or clinic is deploying a new AI diagnostic tool, ask whether it has gone through appropriate validation and whether governance processes (audit, bias review) are in place, rather than assuming vendor claims are sufficient.
5. Stay within the tool’s intended use — using an AI system for a purpose or patient population outside what it was validated for increases both clinical and legal risk meaningfully.

Frequently Asked Questions

If a doctor follows an AI tool’s wrong recommendation, is the doctor automatically protected from liability?

No. The doctor remains responsible for exercising independent clinical judgement; following an AI recommendation without appropriate clinical correlation is generally not a defence and could itself be viewed as a failure to meet the standard of care.

Can a hospital be held liable for deploying a flawed AI diagnostic tool?

Potentially, yes — particularly if the hospital failed to validate the tool appropriately, deployed it outside its intended use, or lacked reasonable governance and oversight processes around its use.

Does India have a specific law governing AI liability in healthcare?

No standalone AI liability law exists as of this writing; liability currently arises through existing frameworks — medical negligence law, the Consumer Protection Act’s product liability and deficiency-in-service provisions, and, in narrow cases, criminal law.

Are fully autonomous AI diagnostic tools (without a doctor’s oversight) legal in India?

Generally no — current standards, including the Telemedicine Practice Guidelines, 2020, do not permit AI platforms to independently diagnose or prescribe without clinician supervision; a doctor is expected to remain the final decision-maker.

What is the single most protective thing a doctor can do when using AI diagnostic tools?

Thorough, contemporaneous documentation of your own independent clinical reasoning — whether you agreed with or overrode the AI’s suggestion — since this is what demonstrates the exercise of genuine clinical judgement rather than passive reliance on the tool.

Researched Sources

1. PMC (National Library of Medicine) — Doctor, Bot, or Both: Questioning the Medicolegal Liability of Artificial Intelligence in Indian Healthcare
2. Dr. Abhishek Gandhi — AI-Based Diagnosis and Liability: Understanding Legal Responsibilities in the Era of Intelligent Healthcare
3. International Bar Association — AI in Healthcare: Trends and Challenges in India
4. International Journal for Multidisciplinary Research — Legal and Ethical Challenges of AI-Based Diagnostics

Disclaimer

This article is for general informational and educational purposes and reflects the legal landscape around AI diagnostic liability as understood at the time of writing; this is a rapidly evolving and largely unsettled area of Indian law. It is not legal advice; doctors and hospitals should consult a qualified healthcare and technology lawyer to assess their specific liability exposure when deploying or using AI clinical tools.