1. Introduction

Artificial Intelligence (AI)-assisted diagnosis liability concerns the civil and legal responsibility that may arise when an AI system used in medical diagnosis contributes to a patient's injury, delayed treatment, incorrect diagnosis, unnecessary medical procedures, or failure to detect a serious disease.

AI-assisted diagnostic technologies are increasingly used in healthcare to interpret radiological images, detect cancer, assess cardiovascular risks, analyse laboratory results, identify neurological disorders, and support clinical decision-making. Although these systems may improve diagnostic accuracy and efficiency, they also create complex questions about medical negligence, product liability, professional responsibility, informed consent, privacy, and compensation.

For example, suppose a hospital uses an AI system to examine a patient's scan. The software incorrectly reports that there is no evidence of cancer. The treating doctor relies on the result without reviewing the scan carefully, and the cancer is diagnosed several months later, after it has progressed. The patient may have a civil claim against the doctor, hospital, software developer, or more than one of these parties, depending on the applicable law and the evidence.

The central legal question is: Who should bear responsibility when an AI-assisted medical diagnosis causes harm, and what must an injured patient prove to obtain compensation?

An important qualification is that most existing medical negligence and product liability cases were decided before modern generative AI and diagnostic algorithms became widespread. The cases discussed below are therefore principally analogous precedents: they establish legal principles that courts may apply to AI-assisted diagnosis, rather than decisions that necessarily concern AI itself.

2. Meaning and scope of AI-assisted diagnosis liability

AI-assisted diagnosis refers to the use of computational systems that analyse medical information and provide predictions, classifications, risk assessments, or diagnostic recommendations to healthcare professionals.

Common examples include:

Radiology AI: detecting abnormalities in X-rays, CT scans, and MRI images.

Pathology AI: identifying suspicious cells or tissue patterns.

Clinical decision-support systems: recommending possible diagnoses based on symptoms, medical history, and test results.

Predictive diagnostic systems: estimating the likelihood of sepsis, stroke, heart disease, or other conditions.

Remote diagnostic platforms: analysing patient-submitted information through telemedicine services.

Automated triage systems: classifying patients according to the urgency of their symptoms.

AI-assisted diagnosis liability arises when the design, development, deployment, supervision, or use of such a system falls below an applicable legal standard and causes compensable harm.

Importantly, an incorrect AI prediction does not automatically establish legal liability. The claimant must generally establish a legally recognised duty, breach of the applicable standard, causation, and actual damage, subject to the relevant jurisdiction's rules and any applicable statutory liability regime.

3. Principal legal theories of liability

A. Medical negligence

Medical negligence is often the most direct basis for a claim when a healthcare professional relies improperly on an AI-generated result.

A doctor may be negligent if the doctor:

Relies blindly on an AI recommendation despite contradictory symptoms or test results.

Fails to investigate a clinically significant abnormality.

Ignores a system warning or an indication that the AI result is unreliable.

Uses a diagnostic tool outside its validated clinical purpose.

Fails to arrange appropriate follow-up testing.

Does not communicate a material diagnostic uncertainty to the patient.

The relevant standard is generally that of a reasonably competent professional in the circumstances, assessed under the law of the jurisdiction concerned. The use of AI does not necessarily reduce the doctor's existing professional obligations.

Example: An AI system marks a lung scan as low risk, but the patient has persistent symptoms and other clinical evidence suggesting cancer. If the doctor fails to investigate those warning signs, the doctor may be liable even if the software itself malfunctioned or generated a misleading result.

Conversely, a doctor is not necessarily negligent merely because a reasonable clinical decision, made with appropriate care, was later shown to be incorrect.

B. Hospital and healthcare institution liability

Hospitals may face liability for their own institutional failures or, where applicable, for the negligence of employees and other personnel for whom they are legally responsible.

Potential grounds include:

Purchasing diagnostic software without appropriate evaluation.

Deploying a system that has not been adequately validated for the relevant patients or clinical setting.

Failing to train clinicians in the system's limitations.

Failing to maintain, update, or monitor the system.

Creating workflows that encourage uncritical reliance on automated results.

Failing to ensure that abnormal results are reviewed and communicated.

Inadequately investigating known system errors or performance disparities.

Hospital liability depends on the applicable rules of direct negligence, vicarious liability, contractual responsibility, and healthcare regulation. A hospital is not automatically liable for every error made by an independent software supplier.

C. Product liability against the AI developer or manufacturer

A developer or manufacturer may face a product liability claim where a diagnostic system has a legally recognised defect that causes injury.

Potential defects include:

Design defects: the system's architecture or decision process creates unreasonable risks.

Manufacturing or implementation defects: the deployed system differs materially from its intended specifications.

Warning defects: users are not adequately informed of limitations, contraindications, or known risks.

Software defects: coding errors, faulty updates, or inadequate testing produce unsafe outputs.

Data-related defects: inadequate or unrepresentative training or validation data produces foreseeable, clinically significant errors.

Whether a software product qualifies for a particular statutory product liability regime is jurisdiction-dependent. Claims may instead, or additionally, arise under negligence, contract, consumer protection, or other legislation.

A plaintiff ordinarily needs to connect the alleged defect to the injury. A software error that had no effect on treatment or outcome may not establish compensable causation.

D. Negligent design, testing, and deployment

AI systems may perform differently across hospitals, equipment, patient populations, and clinical environments. A model developed using one patient population may be less accurate for another.

Legal issues may arise where a developer or healthcare provider:

Fails to test the system in a reasonably representative population.

Fails to evaluate false-negative and false-positive rates.

Ignores foreseeable data drift or performance degradation.

Uses an algorithm for a purpose beyond its validated capabilities.

Fails to investigate recurring diagnostic errors.

Introduces an update without adequate safety testing.

The existence of a statistical error rate alone does not establish negligence. The legal question is whether the relevant party breached an applicable duty and whether that breach caused the claimant's harm.

E. Informed consent and disclosure

Patients may argue that they were not adequately informed about the use of AI in their diagnosis or about material risks affecting their treatment decision.

Relevant questions include:

Was AI used as a decision-support tool or as a substantially autonomous decision-maker?

Did the patient need to be informed about the system's involvement under applicable law or professional standards?

Was a material uncertainty or limitation concealed?

Would proper disclosure have changed the patient's treatment decision?

Did the failure to disclose cause a legally recognised injury?

There is no universal rule that every use of diagnostic AI must be individually disclosed to every patient. The disclosure obligation depends on the jurisdiction, clinical context, materiality of the risk, consent rules, and applicable healthcare standards.

F. Data protection and confidentiality

AI-assisted diagnosis frequently involves sensitive personal health information. Unlawful collection, disclosure, reuse, or security failures may lead to regulatory action or civil claims.

Possible allegations include:

Unauthorised use of medical records to train an AI model.

Disclosure of identifiable diagnostic information.

Failure to implement appropriate data security.

Improper sharing of medical information with technology vendors.

Breach of applicable privacy or confidentiality duties.

A privacy breach and a diagnostic injury are distinct legal issues. A patient may have a privacy-related claim even without a wrong diagnosis, while a diagnostic negligence claim may succeed without any privacy violation.

4. Essential elements of a civil liability claim

A patient bringing a civil claim will generally need to establish the following elements, although the exact tests and burdens of proof vary by jurisdiction.

1. Duty of care

The defendant owed a legally recognised duty to the patient, such as a doctor's professional duty or a manufacturer's applicable duty concerning product safety.

2. Breach of duty

The defendant failed to meet the relevant professional, institutional, contractual, or product safety standard.

3. Causation

The claimant must establish the required causal connection between the breach and the injury, under the applicable legal test. The fact that AI produced an incorrect result is not enough by itself.

4. Legally recognised damage

The claimant must demonstrate a compensable injury or loss, such as physical harm, additional treatment costs, lost earnings, or another form of damage recognised by the relevant law.

The special problem of causation

Causation can be particularly difficult in diagnostic AI cases because the patient's underlying disease, the AI output, the doctor's decision, and the eventual medical outcome may all contribute to the harm.

For example, if an AI system misses a tumour, the claimant may need to demonstrate that timely detection would probably have led to a materially better outcome under the applicable causation standard.

Where the disease would have caused the same outcome regardless of the error, a claim for the full consequences of that outcome may fail. Some legal systems, however, recognise particular forms of loss associated with lost chances, delayed treatment, or increased risks in specified circumstances.

Courts may consider expert evidence concerning:

The original medical images and clinical records.

The AI system's validated accuracy and known limitations.

The treating doctor's actions and professional standard of care.

Whether a competent clinician should have detected the condition.

The likely course of the disease with timely treatment.

Whether the alleged breach materially contributed to the injury.

5. At least eight important case laws

The following decisions establish principles relevant to AI-assisted diagnosis liability. They are not all AI cases; their importance lies in how their legal reasoning can be applied to disputes involving diagnostic algorithms, clinicians, hospitals, and software providers.

Case 1. Bolam v Friern Hospital Management Committee (1957)

Court: Queen's Bench Division, England

Legal principle: Professional negligence and the standard of reasonable medical practice.

Facts: A patient undergoing electroconvulsive therapy suffered injuries. The case examined whether the medical professionals had acted negligently by not using muscle relaxants and by not providing additional precautions.

Decision: The court formulated the principle commonly known as the Bolam test: a medical professional may avoid a finding of negligence if acting in accordance with a practice accepted as proper by a responsible body of medical professionals skilled in that field, subject to subsequent judicial qualifications.

Application to AI-assisted diagnosis: A doctor who uses AI diagnostic software must be assessed against the applicable standard of competent medical practice. Relevant considerations may include whether responsible clinical practice supports the use of that tool, whether it was used for an appropriate purpose, and whether the clinician exercised adequate independent judgment.

However, merely showing that some doctors use a particular AI system does not automatically establish that the practice is reasonable or lawful. The Bolam principle must be applied with the qualifications recognised in later case law and the governing jurisdiction.

Significance: This case provides a starting point for determining whether a clinician's reliance on diagnostic AI amounts to professional negligence.

Case 2. Bolitho v City and Hackney Health Authority (1997)

Court: House of Lords, United Kingdom

Legal principle: Professional medical opinion must withstand logical analysis.

Facts: A young child suffered respiratory distress. A doctor failed to attend when requested, and the dispute concerned whether proper attendance would have resulted in intubation and prevented the injury.

Decision: The House of Lords held that a court is not required to accept professional medical opinion merely because a responsible group of professionals supports it. The opinion must have a logical basis and be capable of withstanding rational scrutiny.

Application to AI-assisted diagnosis: A hospital cannot necessarily defend an AI-based diagnostic process simply by asserting that its clinicians or software vendor regarded the system as reliable.

A court may examine whether reliance on the system was logically justified in the particular circumstances. For example, if the AI system was known to perform poorly on a particular type of scan, or if its output contradicted strong clinical evidence, uncritical reliance may be difficult to justify.

Significance: This decision is especially relevant where a defendant attempts to justify an AI-assisted clinical decision by reference to accepted practice without adequately explaining its safety or logic.

Case 3. Montgomery v Lanarkshire Health Board (2015)

Court: Supreme Court of the United Kingdom

Legal principle: Disclosure of material risks and patient autonomy.

Facts: A pregnant patient with diabetes was not adequately warned about the material risks associated with vaginal delivery, including the risk of shoulder dystocia and injury to the baby.

Decision: The Supreme Court emphasised the patient's right to make informed decisions. A doctor generally must take reasonable care to ensure that the patient is aware of material risks involved in a recommended treatment and of reasonable alternative or variant treatments.

Application to AI-assisted diagnosis: The decision is relevant where AI affects a patient's choice between diagnostic or treatment options. If the use of an AI system creates a material risk or uncertainty that should be disclosed under the applicable law, failing to communicate it may undermine informed consent.

For instance, a patient deciding between conventional diagnostic review and an AI-supported procedure may need information about clinically material differences in risks, reliability, or available alternatives.

Important qualification: Montgomery does not establish a universal obligation to disclose every software tool or algorithm. Whether disclosure is required depends on the materiality of the information and the circumstances of the case.

Significance: The case helps explain why diagnostic AI must be considered in the broader context of patient autonomy, informed consent, and transparency.

Case 4. Jacob Mathew v State of Punjab (2005)

Court: Supreme Court of India

Legal principle: Medical negligence and the standard expected of medical professionals.

Facts: The case concerned allegations of medical negligence following a patient's death in a hospital, including issues relating to the medical care provided and the equipment available.

Decision: The Supreme Court discussed the standard of care applicable to medical professionals and distinguished ordinary errors of judgment from actionable negligence. It emphasised that negligence requires more than an adverse medical outcome or a mere error.

The judgment primarily addressed criminal medical negligence, but its discussion of professional competence is also relevant to understanding the wider law of medical responsibility. Civil claims must still be assessed under the applicable civil standards.

Application to AI-assisted diagnosis: A doctor is not automatically negligent merely because an AI recommendation turns out to be wrong. The central question is whether the doctor failed to exercise the required degree of professional care.

Equally, a doctor cannot necessarily escape liability by arguing that an AI system supplied the incorrect answer. If the doctor failed to verify an obviously questionable result or ignored important symptoms, the clinician's conduct must be independently assessed.

Significance: In India, this judgment is a useful starting point for distinguishing genuine negligence from a reasonable clinical judgment that produces an unfortunate outcome.

Case 5. Kusum Sharma v Batra Hospital and Medical Research Centre (2010)

Court: Supreme Court of India

Legal principle: The standard of care, professional judgment, and the distinction between negligence and an unsuccessful treatment outcome.

Facts: The case arose from allegations of medical negligence concerning hospital treatment and the death of a patient.

Decision: The Supreme Court examined the principles governing medical negligence and emphasised that a doctor should not be held negligent simply because the treatment was unsuccessful or another approach might have been preferable. The professional's conduct must be judged against the applicable standard of reasonable competence and care.

Application to AI-assisted diagnosis: Suppose a doctor uses a properly validated diagnostic system, considers the patient's symptoms, evaluates relevant test results, and makes a reasonable decision, but the disease is nevertheless missed. The adverse outcome alone would not establish negligence.

On the other hand, a clinician who relies mechanically on an algorithm despite obvious warning signs may fall below the applicable standard.

The same distinction applies to hospitals choosing AI tools: a later failure does not automatically prove negligence, but inadequate evaluation, training, or safeguards may support a claim where they caused harm.

Significance: This case is particularly useful in Indian civil litigation involving allegations that a healthcare provider improperly relied on AI.

Case 6. Samira Kohli v Dr. Prabha Manchanda (2008)

Court: Supreme Court of India

Legal principle: Informed consent and the limits of consent to medical procedures.

Facts: A patient consented to a diagnostic procedure, but a more extensive surgical intervention was performed without the patient's specific consent in circumstances that generated a legal dispute.

Decision: The Supreme Court explained that valid consent requires adequate information and voluntary authorisation for the procedure in question. Consent to one intervention does not ordinarily amount to unrestricted permission to perform a substantially different procedure, subject to recognised exceptions such as genuine emergencies.

Application to AI-assisted diagnosis: The case is relevant when AI is incorporated into diagnostic procedures or influences the nature of an intervention. A patient's consent to a medical procedure should not automatically be treated as consent to every possible intervention or materially different procedure that an AI system might recommend.

Whether separate disclosure or consent is required for AI itself is a distinct question and depends on the applicable law and clinical context.

Significance: This decision supports patient autonomy and reinforces the need to examine what the patient actually authorised, particularly where AI recommendations lead to additional procedures.

Case 7. Rogers v Whitaker (1992)

Court: High Court of Australia

Legal principle: Disclosure of material risks and the patient's right to make an informed decision.

Facts: A patient who had limited vision in one eye underwent an eye operation. A rare complication affected the previously healthy eye, causing serious consequences. The surgeon had not adequately warned the patient of that risk.

Decision: The High Court of Australia rejected an exclusively profession-centred approach to disclosure and held that a doctor must disclose material risks. A risk may be material where a reasonable person in the patient's position would likely attach significance to it, or where the doctor knows or should know that the particular patient would likely attach significance to it.

Application to AI-assisted diagnosis: If a diagnostic AI system has a known limitation that materially affects the safety of a proposed procedure, the relevant healthcare professional may need to disclose that risk, depending on the circumstances and governing law.

For example, if a system has a known tendency to miss certain abnormalities in a relevant patient group, that limitation may matter when the patient is deciding whether to rely on the diagnostic process or pursue further testing.

Significance: This case illustrates why the patient's perspective and the materiality of a diagnostic risk may matter more than a general assurance that the AI system is widely used.

Case 8. Donoghue v Stevenson (1932)

Court: House of Lords, United Kingdom

Legal principle: Duty of care in negligence and the protection of persons foreseeably harmed by a defective product.

Facts: A consumer alleged illness after consuming ginger beer that allegedly contained a decomposed snail. The product had been purchased by a friend, and there was no direct contract between the consumer and the manufacturer.

Decision: The House of Lords recognised the manufacturer's duty of care to the ultimate consumer in the circumstances of the case. The judgment became foundational to modern negligence law and the neighbour principle.

Application to AI-assisted diagnosis: The case provides a historical foundation for considering whether an AI developer or manufacturer owes duties to patients who may foreseeably be harmed by an unsafe diagnostic system.

Depending on the governing law, a patient may have a claim even without a direct contract with the developer. The exact scope of the duty, however, depends on the applicable negligence and product liability rules.

Significance: This decision is relevant when identifying potential responsibility beyond the immediate doctor-patient relationship, especially where a software developer's conduct may have contributed to the injury.

Case 9. V. Kishan Rao v Nikhil Super Speciality Hospital (2010)

Court: Supreme Court of India

Legal principle: Consumer protection and proof of medical negligence.

Facts: The case concerned allegations of medical negligence in the treatment of a patient who was suspected of suffering from malaria. The dispute included the question of whether expert evidence was invariably necessary to establish medical negligence.

Decision: The Supreme Court clarified that expert evidence is not mandatory in every medical negligence case. Where the facts themselves adequately demonstrate negligence, a consumer forum may decide the matter without insisting on expert testimony in every instance.

Application to AI-assisted diagnosis: Consider a situation in which a hospital's records show that an AI system flagged a potentially serious abnormality, but the hospital's workflow failed to communicate the alert to the treating doctor. If the evidence clearly demonstrates an avoidable failure, expert testimony may not always be indispensable to establishing that particular breach.

However, determining whether a complex AI model was defective or whether a missed diagnosis caused a particular medical outcome will often require specialist evidence.

Significance: This decision is relevant to Indian patients pursuing claims before consumer commissions when an AI-related failure appears from the available medical records.

Case 10. Chester v Afshar (2004)

Court: House of Lords, United Kingdom

Legal principle: Informed consent and causation.

Facts: A patient underwent spinal surgery without being adequately warned of a small but serious risk of neurological injury. The risk materialised.

Decision: The House of Lords allowed the patient's claim in the particular circumstances, addressing the relationship between the failure to warn of a material risk and the injury that occurred.

Application to AI-assisted diagnosis: The case may be relevant where a patient alleges that failure to disclose a material diagnostic risk prevented an informed decision about whether to undergo a procedure, seek a second opinion, or pursue further testing.

The claimant would still need to satisfy the relevant jurisdiction's causation requirements. Chester is a particular decision concerning failure to warn, not a general rule that every AI-related disclosure failure establishes liability.

Significance: It demonstrates that causation in medical negligence cases may involve complex questions about patient choice, not merely the technical accuracy of a diagnostic output.

6. How liability may be distributed among different parties

An AI-assisted diagnosis can involve several actors. Responsibility depends on what each party did, the duties owed, and the connection between the conduct and the injury.

Several parties may be responsible for the same injury. The applicable jurisdiction determines whether liability is joint, several, proportionate, or otherwise allocated.

A contract between a hospital and an AI developer may allocate costs and responsibilities between them. Such an agreement does not automatically eliminate a patient's independent legal rights.

7. Illustrative case study: AI fails to detect cancer

Consider the following hypothetical situation.

A patient undergoes a CT scan at a private hospital. The hospital uses AI software to assist with the interpretation. The software reports no significant abnormality, and the treating doctor accepts the result without further investigation. Six months later, the patient is diagnosed with advanced cancer.

The patient brings a civil claim.

Potential legal analysis

Issue 1 — Doctor's negligence

Did the doctor adequately consider the symptoms, scan findings, and other available clinical information, or rely unreasonably on the AI output?

Issue 2 — Hospital's negligence

Did the hospital implement appropriate validation, training, escalation procedures, and human review?