Author: Ruchi Kol
College: University of Allahabad Faculty of Law
Abstract:- Artificial Intelligence is no longer just assisting professionals; it is actively taking decisions in high-stake fields like medicine and law. But when an AI system gives a wrong diagnosis or drafts a faulty legal contract, who bears the liability? This article examines the emerging legal vacuum around AI liability in India and other jurisdictions.
Keywords:- Artificial Intelligence, Medical Negligence, Legal Tech, Liability, Tort Law, Consumer Protection
Introduction
In early 2023, a US lawyer was fined for submitting a legal brief drafted by ChatGPT that cited six completely fake cases. In the same year, a Belgian man reportedly died by suicide after an AI chatbot encouraged him to do so. These are not isolated incidents anymore.
Today, hospitals in Delhi and Mumbai are using AI tools like IBM Watson for oncology and http://Qure.ai for reading X-rays. Similarly, law firms are using AI platforms like Harvey and CaseMine to draft contracts and predict case outcomes. The efficiency is undeniable, but it raises a critical legal question that our current laws are not prepared to answer – if the AI makes a mistake, who is legally responsible?
Is it the doctor who trusted the AI? The hospital that bought it? The company that developed it? Or the AI itself?
1. The Traditional Framework of Liability and Why It Fails for AI*
Under Indian law, liability is generally fixed on two principles:
a) Negligence (Law of Torts) :- To prove negligence, one must prove duty, breach, and damage. In a medical negligence case under the classic _Jacob Mathew v. State of Punjab (2005)_ standard, a doctor is expected to exercise reasonable care. But what if the doctor exercises reasonable care and relies on an AI system approved by the hospital, and the AI itself is wrong? The doctor hasn’t been negligent in the traditional sense.
b) Product Liability (Consumer Protection Act, 2019):-Section 2(34) of the Act defines product liability. If a product has a manufacturing defect or faulty design, the manufacturer is liable. Can we treat an AI software as a “product”? Arguably yes. But AI is not static. It learns and evolves after deployment. A defect may not exist at the time of manufacturing but may develop due to biased training data. This makes it very difficult to fix liability under traditional product liability.
2. Medical AI Errors: A Case Study
Imagine a scenario: An AI radiology tool fails to detect early-stage lung cancer in a patient. The radiologist, relying on the AI’s “All Clear” report, discharges the patient. Six months later, the cancer reaches Stage 4.
Who is at fault here?
There are three possible approaches courts worldwide are considering:
i) The Doctor/Hospital is liable :- The most likely outcome in India today. Under the principle of _respondeat superior_, the hospital is vicariously liable for its tools and staff. The law will say AI is merely an assistive tool, like a stethoscope. The final decision and therefore the final liability rests with the human professional. This is the view adopted by the US FDA as well – AI is a “Clinical Decision Support System”, not a replacement for a doctor.
ii) The Developer is liable :- If it can be proven that the AI was trained on inadequate or biased data, or that the algorithm itself was flawed, the developer company can be held liable under product liability. For instance, if an AI trained mostly on data of foreign patients fails to accurately diagnose Indian patients, it is a design defect.
iii) Shared Liability Model :- This is the most practical model. The liability should be apportioned. The developer for creating a faulty system, the hospital for not properly validating it before deployment, and the doctor for over-relying on it without applying independent clinical judgment.
3. Legal AI Errors: A Different Challenge
Legal errors caused by AI have a different nature. In medicine, the damage is physical. In law, the damage is financial and reputational.
If an AI drafts a contract with a wrong indemnity clause and the client loses crores, can the client sue the AI company? Under the Advocates Act, 1961, only an advocate enrolled with the Bar Council can practice law. An AI cannot be an advocate. Therefore, a client always hires a lawyer, not the AI. The lawyer cannot escape liability by saying “AI ne galti ki”. Under Section 26 of the Bharatiya Nyaya Sanhita and under professional misconduct rules, the advocate remains fully responsible for all work submitted under his name.
The famous _Mata v. Avianca, Inc. (2023)_ case in the US Federal Court is a perfect example. The court did not punish ChatGPT; it punished the lawyers who failed to verify the information.
4. What Does Indian Law Say?
India does not yet have a dedicated AI liability law. The NITI Aayog’s National Strategy for Artificial Intelligence (2018) and the proposed Digital India Act are silent on the question of liability. Currently, we have to rely on a patchwork of laws:
. Information Technology Act, 2000 (Sec 43A & 79):-Deals with intermediary liability, but not sufficient for AI autonomy.
. Consumer Protection Act, 2019 :- Can be stretched to cover AI as a service/product.
. Bharatiya Nyaya Sanhita, 2023 :- If an AI error causes death due to negligence, Section 106 (causing death by negligence) could potentially be invoked against the human operator.
The European Union is far ahead with its EU AI Act, 2024, which classifies medical and legal AI as “High-Risk AI Systems” and imposes strict obligations on developers for transparency, human oversight, and quality of training data.
Conclusion: The Way Forward
We cannot apply 19th-century tort principles to 21st-century autonomous systems. The concept of AI as a separate legal person, as once proposed for Sophia the robot in Saudi Arabia, is still science fiction and legally unsound because an AI has no assets to pay compensation.
For India, a three-step approach is needed:
1. Mandatory Human-in-the-Loop :- The law must it compulsory that no AI decision in high-risk areas like medicine and law can be executed without meaningful human review. The human must remain the final decision-maker.
2. AI Insurance :- Just like motor vehicle insurance, we need mandatory AI liability insurance. If an AI tool fails, the insurance should compensate the victim, similar to the no-fault liability model.
3. Clear Disclosure and Consent :- Hospitals and law firms must inform the client/patient that AI is being used and take their informed consent. This is already a part of medical ethics.
Ultimately, AI does not erase accountability; it redistributes it. Until the law catches up, the safest legal position remains – the person who chooses to use the AI tool will be the person who has to answer for its mistakes. Technology can be artificial, but liability will always be human.
FAQS
Q1. Can AI be sued in India?
No. AI is not a legal person. Liability is always on the human – doctor, lawyer, hospital or developer.
Q2. Who is liable if AI gives a wrong medical diagnosis?
The doctor/hospital is primarily liable, as AI is considered only an assistive tool. Developer can also be liable if the AI had biased data or design defect.
Q3. What if a lawyer uses AI and it makes a legal error?
The lawyer is fully liable under Advocates Act, 1961. He cannot say “AI made the mistake”. Court will punish the lawyer for not verifying.
