AppsGuidesAI DictionarySmall FixesOn RecordAboutFollow the lab

AI liability

Also called who is responsible when ai is wrong

AI liability is the question of who has to pay or answer in law when an AI system causes harm: the maker, the business using it, or the user.

Safety and societyCurious

In plain words

If a pressure cooker explodes, the law has long had answers: the maker may be at fault, or the shop, or the cook who ignored the instructions. AI liability asks the same question about AI. A chatbot gives wrong medical advice, a hiring tool rejects someone unfairly, a self-driving car crashes. Who is responsible?

It is harder than it sounds. Many hands touch an AI system: one company trains the model, another builds it into an app, and a person uses it in their own way. AI can also behave in ways nobody predicted, which makes it hard to prove what went wrong.

So far, courts have mostly used old laws: negligence, consumer protection and product liability. One early answer is clear, though. A business cannot blame its chatbot as if it were a separate person.

Everyday example

A wrong medicine from the chemist

You ask the chemist for your usual tablets and get the wrong strength. Was it the drug company's labelling, the chemist's mistake, or did you misread the prescription? The answer decides who pays. With AI, the model maker is the drug company, the business using it is the chemist, and you are the patient.

Where the example breaks: A medicine either matches its label or it does not. An AI's output can change each time, and the reason for a bad answer may be buried in billions of numbers, so proving a defect is much harder.

Go deeper

How it actually works

Liability rules decide who carries the cost of harm, and so who has a reason to prevent it. Here is how the main approaches stood as of November 2026.

  1. The chatbot is part of the business

    In Moffatt v. Air Canada (February 2024), the airline's website chatbot wrongly told a grieving customer he could claim a bereavement discount after travelling. Air Canada argued the chatbot was a separate legal entity. A Canadian tribunal disagreed: the airline was responsible for all information on its website and paid damages.

  2. Fault versus strict liability

    Under negligence, the injured person must show someone was careless. Under strict liability, a maker pays for a defective product even without proof of carelessness. Which rule applies to AI, and whether AI counts as a product or a service, is the central debate.

  3. EU: software becomes a product

    The EU dropped its proposed AI Liability Directive, formally withdrawn in October 2025. Instead, the new Product Liability Directive, applying to products placed on the market after 9 December 2026, counts software as a product. It keeps strict liability and lets courts presume a defect when technical complexity makes proof excessively hard.

  4. United States

    There is no federal AI liability law. Cases go through ordinary courts under product liability, negligence and consumer law, and judges are still deciding how they apply to chatbots and AI tools.

  5. India

    India has no AI-specific liability law as of November 2026. The Consumer Protection Act, 2019, with its "deficient service" and product liability rules, is the likeliest route. An IIT Madras working paper (2026) found its definitions wide enough to cover AI harms, but proving defect and cause would be hard.

Landmark case
Moffatt v. Air Canada, British Columbia Civil Resolution Tribunal, 14 February 2024
EU
Product Liability Directive (EU) 2024/2853 applies to products placed on the market after 9 December 2026; AI Liability Directive withdrawn (Official Journal notice, 6 October 2025)
India
No dedicated law as of November 2026; Consumer Protection Act 2019 and the IT Act apply. Platform safe harbour may not clearly cover AI-generated content, per the 2025 AI Governance Guidelines
Not legal advice
We are not lawyers. Laws differ by country and change often

Should AI makers be strictly liable for harm?

One view

Yes, put cost on the maker

Users cannot see inside a model or prove what went wrong. Making makers pay for defects, as with other products, gives them a strong reason to test and fix systems before release.

The other view

No, share it by role

A general model can be used in countless ways its maker never saw. Heavy liability could chill open research and small developers. Responsibility should sit mostly with whoever deploys AI in a risky setting.

Most likely: Most proposals end up in between: makers answer for defects in what they ship, businesses answer for how they deploy it, and courts get help with evidence. The EU has chosen strict product liability for software; most other places, including India and the US, are still working it out.

Questions people ask

Who is responsible when a chatbot gives wrong information?

Usually the business that offers the chatbot to its customers. In the 2024 Air Canada case, a tribunal ruled the chatbot was part of the airline's website, so the airline had to pay. The model maker may also be liable in some cases, depending on contracts and local law.

Can I sue an AI?

In practice, no. Courts treat AI as a tool, not a legal person. When Air Canada argued its chatbot was a separate legal entity, the tribunal rejected the idea. Claims are brought against the companies or people who make, sell or use the AI.

What was the EU AI Liability Directive?

A 2022 European Commission proposal to make it easier to claim compensation for harm caused by AI. It was withdrawn in 2025. The updated Product Liability Directive, applying from December 2026, now covers software, including AI.

Does India have an AI liability law?

Not a dedicated one, as of November 2026. People harmed by AI would rely on existing laws such as the Consumer Protection Act, 2019, the IT Act and contract law. Legal experts say who is liable, and how much, remains unclear.

Is the AI Act the same as AI liability law?

No. The EU AI Act sets rules for building and using AI safely. Liability law decides who pays compensation after harm. They connect, because breaking safety rules can help prove fault.

Was this clear?

1 is confusing, 5 is crystal clear. One tap, no account needed.

Spotted something wrong, or know a better example? Suggest a change

Further reading

Prepared by Mundhra Labs. Reviewed by Aditya Mundhra. Checked November 2026.