AI liability

advancedspreading

AI liability addresses the question of who must legally stand accountable and pay if an artificial intelligence system causes harm to someone.

In plain terms

If a dog bites someone, the owner is sued, not the dog. AI liability raises the same question for machines: if an AI system makes a mistake and causes damage, who is the 'owner'? The developer who wrote it, the company that deployed it, or the person using it?

The answer is difficult because a long chain stands behind an AI: someone trained the model, another built a product on it, a third company configured it and presented it to clients. Moreover, the system's operation is often opaque from the outside, making it hard for the victim to prove where the error occurred.

AI liability is not a single law but a set of rules: old tort law, product liability, contracts, and new regulations specifically tailored for AI. The goal is twofold: the victim should not be left without compensation, and companies should know in advance what they are liable for.

How it works

First, it must be clarified whether harm occurred and what kind: financial loss, physical injury, destroyed data or, for example, discrimination.
Afterwards, they examine who did what in the chain: who built the system, who operated it, who gave instructions, and whether everyone complied with the applicable rules.
Law recognizes two types of logic. In one, negligence must be proven (someone did not act as expected); in the other, it is enough that the product was defective and damage resulted.
Since the victim cannot inspect the system, new EU product liability rules assist them: courts can compel companies to release evidence and, in certain cases, presume the product was defective or that the defect caused the harm.
The same principle applies to autonomous agents: the agent is not a legal entity, so the person who operates it or allows it to act on their behalf is responsible for its actions.

An everyday example

An airline chatbot gives you incorrect information about ticket refunds, and you buy a ticket based on this. A Canadian small-claims civil dispute resolution body ruled in 2024 in exactly such a case (Moffatt v. Air Canada) that the company is liable for its robot's statements and cannot claim "the chatbot said it."

Who is affected

entrepreneur, executiveIf your company allows AI or an agent near customers, the company is likely liable for errors, not the software provider.

developerUnder the new EU product liability directive, software, including AI systems, is considered a product, so the manufacturer can be held liable for faulty operation. This applies to products placed on the market after December 9, 2026, and does not apply to free, open-source software developed outside of business activities. → Code

lawyer, citizenAs a victim, you need to know whom to contact, and that the new EU product liability rules make proving liability easier. → Law

patient, doctorIf AI assists the diagnosis, the doctor and the institution typically remain responsible for the final decision. → Health

A common misconception

MythIf the AI made a mistake, no one is to blame since the machine decided.

In realityAI is not a legal entity, so responsibility cannot be shifted to it. The responsible party must always be sought among people and companies: the manufacturer, the operator, or the user may be liable for the damage.

Limits and risks

Rules vary by country, and on many issues there is no established judicial practice yet, so the outcome of a specific case is difficult to predict.
Terms of use often shift responsibility to the user, so it is worth reading them before entrusting an important task to an AI.
If you forward or use an AI's response without verification, you may also be held responsible for errors.
With autonomous agents, the error chain is long, making it hard to determine later which step and whose negligence caused the damage.

Similar, but not the same

Accountability – Accountability is a broader organizational and ethical principle regarding who must provide an explanation, whereas AI liability refers to court-enforceable compensation obligations.
AI compliance – Compliance means adhering to regulations in advance, while liability refers to who pays afterward if damage occurs.
AI safety – AI safety focuses on preventing harm, while AI liability deals with the legal resolution of harm that has already occurred.

Related concepts

Broader

AI regulation, AI governance, Right to compensation

Narrower

Product liability, Burden of proof, strict liability

On the edge

EU AI-rendelet, Accountability, Human oversight, Explainability, AI compliance

A brief history

In 2022, the European Commission proposed a separate AI liability directive, but withdrew it in 2025. Meanwhile, in 2024, the new EU Product Liability Directive was adopted, replacing the 1985 regulation and treating software, including AI systems, as products.

Where things stand

EU member states must transpose the new EU Product Liability Directive by December 9, 2026, and it applies to products placed on the market thereafter. There is no separate EU AI liability law; courts decide disputed cases largely based on existing compensation rules.

Status: 2026 November

In the news

Appears in 21 Agent Age stories; the curve shows weekly occurrence in the news magazine.

AI liability in insurance: AI law firm Crosby buys malpractice cover as risk migrates to cyber insuranceNov 4, 2026 · Money
New York City clarifies employer liability for AI-assisted hiring decisionsNov 2, 2026 · Work & HR
Clinical AI liability: AMA briefing and legal risks for health systemsOct 27, 2026 · Jog
AI liability becomes more pressing as insurers and companies rethink coverageOct 26, 2026 · Money
AI liability rules: EU centrists clash while England and Wales law commission proposes reformOct 26, 2026 · Jog
Law Commission outlines plan to overhaul product liability for AI, apps and algorithmsOct 22, 2026 · Jog
AI exclusions spread in insurance; Harbor raises $10 million to cover the gapOct 19, 2026 · Money
Insurers weigh AI exclusions as Marsh chief warns of uncertaintyOct 15, 2026 · Money

Also known as

AI liability · AI liability · artificial intelligence liability · AI liability · AI liability · liability for AI

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.