AI agent liability depends on who suffered which damage and what role your business holds. Review the applicable law alongside your supplier and customer contracts.
What you'll learn
- Distinguish regulatory complaints from compensation claims.
- Check the claimant, damage type and your legal role.
- Review contracts, insurance and action controls together.
An AI agent does not take over your business’s legal responsibilities. The AI Act creates regulatory duties, while the revised Product Liability Directive covers specified damage suffered by natural persons. The separate AI Liability Directive proposal was withdrawn. For a wrong invoice, a missed booking or another business loss, check national liability rules, other applicable laws and both sides of your contracts before granting the agent authority.
The question "who is liable when AI makes a mistake" returns a large amount of published debate about whether a machine can hold legal responsibility. Almost none of it tells a European business what happens to them when the agent they deployed sends the wrong invoice, books the wrong slot, or tells a client something untrue.
This page answers the narrower and more useful version. The AI Act, revised product liability rules and withdrawn liability proposal concern different parts of the problem. Start with their scope, then look at the contracts and other applicable rules.
For a specific claim, ask a lawyer in the relevant country. Identify the person who suffered the loss, the kind of damage, the product’s date and your business’s role. Those facts determine which routes deserve investigation.
Does the EU AI Act compensate anyone harmed by an AI agent?
Article 85 of the AI Act provides a complaint route to a market surveillance authority, not a civil damages award. A regulatory penalty and compensation for your customer are separate questions.[1]
Article 85 gives "any natural or legal person" who suspects an infringement the right to complain to a market surveillance authority.[1] That is a right to have the authority look. It is not a right to be paid. The article creates no entitlement to compensation, restitution, or damages.
This distinction gets lost constantly, and it matters commercially. If your agent breaches an AI Act obligation and a regulator fines you, that fine does not compensate the client who was harmed. Your client's claim against you is a separate matter, running on a separate track, under different law. You can face both at once.
A regulatory fine punishes you. It does not pay the person your agent harmed. Those are two different exposures, and settling one does not settle the other.
The EU AI Act deployer guide covers what the Act does require: which role you hold, which transparency duties attach, and what the intake looks like.
Does the Product Liability Directive cover AI systems?
Yes. Directive (EU) 2024/2853 expressly includes software in Article 4(1). That clarifies the treatment of AI products; it does not mean software was previously outside every product liability rule. The Commission says the earlier regime also covered software.[2][3]
The claimant does not have to prove negligence. Article 10 starts from defectiveness, damage and a causal link, with conditional disclosure and rebuttable presumptions under Articles 9–10 to help with evidential difficulties. A bad result alone does not automatically make the supplier liable.[2]
The national transposition deadline is 9 December 2026. Article 2(1) places products put on the market or into service after that date within the revised directive’s scope. The earlier regime continues for earlier products; check the relevant national implementing law and its transition rules.[2][6]
Then come the limits, and for a business they are severe.
Can a company claim under the Product Liability Directive?
Not for the company’s own loss under this directive. Article 5(1) protects an injured natural person. Article 5(2) also permits claims by someone who succeeds to or is subrogated to that person’s right, or acts on their behalf. A company’s own loss and an acquired claim are different situations.[2]
The heads of damage in Article 6(1) narrow it further:[2]
| Covered | Not covered |
|---|---|
| Death and personal injury, including medically recognised psychological harm | Damage to property used exclusively for professional purposes (Art. 6(1)(b)(iii)) |
| Damage to property used for private or mixed purposes | Destruction or corruption of data used for professional purposes |
| Destruction or corruption of data not used for professional purposes | Pure economic loss, privacy infringement, discrimination (Recital 24) |
Read the middle row carefully, because the wording is more forgiving than the summaries suggest. Property used for mixed purposes is covered. Only property used exclusively for professional purposes falls outside. A laptop that runs the business and also the household is not automatically outside the Directive.
Recital 24 says pure economic loss, privacy infringements and discrimination do not by themselves trigger liability under this directive. Article 6(2), however, covers material losses resulting from the listed damage, and qualifying non-material losses under national law. Do not confuse a stand-alone trading loss with financial consequences of a covered injury.[2]
Say an agent misses a booking and the business loses the fee, without injury or covered property damage. That example points toward contractual and other national remedies rather than this directive. A sole trader is still a natural person, but professional property and professional data remain subject to the stated exclusions. Ask about the actual claimant and loss, rather than relying on the label “business”.
One asymmetry cuts the other way. Article 15 says that liability under the Directive cannot be limited or excluded by contract, or by national law, as against the injured person.[2] So where the Directive applies, a supplier cannot contract its way out. That protection runs to the injured natural person. It does not run to your business.
The instrument that finally made software a product is written for the person your agent harms, not for the business that deployed it.
Who is liable under the Directive, and is it the business running the agent?
The Directive points liability at a cascade of economic operators.[2] The list runs: manufacturers of a finished product or component. Anyone who substantially modifies a product after it goes on the market. Importers and authorised representatives, where the manufacturer sits outside the EU. Fulfilment service providers. And in defined circumstances, distributors and online platforms.
“Deployer” is not a complete answer to the liability question. Article 4(10) includes a person who develops or manufactures a product for its own use in the manufacturer definition. Article 8(2) also addresses a substantial modifier who subsequently makes the product available or puts it into service. Work out whether your business is merely using a supplier’s product or is also developing or substantially modifying it.[2]
For substantial modification, Article 4(18) supplies a definition tied to relevant safety rules or, where they are absent, changes to original performance, purpose or type and the hazard or risk. Article 8(2) adds the condition that the change happens outside the original manufacturer’s control. Keep the change specification and allocation of responsibility; do not assume every configuration change makes you a manufacturer, or that own-use deployment always keeps you outside the rules.[2]
What happened to the AI Liability Directive?
The Commission withdrew the proposed AI Liability Directive on October 6, 2025. The European Parliament’s procedure record marks the proposal as lapsed or withdrawn; it did not become a damages regime.[4]
The proposal aimed to ease evidential difficulties in fault-based AI claims. Its withdrawal does not remove existing contract, tort or other statutory remedies. It means you should not negotiate on the assumption that this proposed EU-wide route already protects you.
For a contract discussion, separate the live rules from proposed changes. Ask your adviser which national causes of action apply, what evidence you need to preserve and whether any other sector-specific or data protection law matters. Do not wait for a successor proposal before deciding who handles a customer complaint or funds the first response.
Is the Product Liability Directive actually in force across Europe yet?
The directive entered into force on December 8, 2024, but member states have until December 9, 2026 to implement it. Entry into force, national implementation and applicability to a particular product are separate steps.[6]
Germany’s Bundestag adopted its implementation bill on October 8, 2026. That parliamentary decision is a current milestone, not a statement that the new liability regime already applies to your product. Check the enacted national text, commencement and transition provisions before relying on it.[5]
For a cross-border service, record where a claim could arise and ask which law applies to the relevant contract and loss. A country’s implementation bill is more useful than a general headline saying Europe has adopted new rules. Keep the product’s supply and deployment dates with that advice, because the transition can change the applicable regime.
The directive harmonises the matters within its scope, subject to the exceptions it expressly permits. Article 18 provides a route for national measures concerning the development-risk defence; that is a specific permitted difference, not permission for unrestricted national divergence.[2]
So what actually governs when your agent harms a client?
Start with the actual loss and legal role. The AI Act’s complaint route, the revised product liability rules and the withdrawn proposal answer different questions. None is a complete map of every claim your business might face.[1][2][4]
National contract and non-contractual liability remain central to a business-loss claim. Other regimes can also apply: the directive expressly preserves EU data protection law and other national liability routes in Article 2(4). The right question for your adviser is which routes fit this claimant and this damage, rather than whether “AI liability” exists as one separate category.[2]
That is not a bad outcome for a business that reads its contracts. It is a very bad one for a business that has not.
Practically, three documents decide your exposure long before any court does:
- Your provider's terms. Look for the liability cap, what it is a multiple of, and what it excludes. Compare the cap with the loss you could face; a fee-based cap can leave a gap. Check whether indemnities cover third-party claims from your clients.
- Your own client contracts. If your agent touches client work, your liability to that client is governed by what you agreed with them, not by what your provider agreed with you. A gap between the two is a gap you fund.
- Your insurance. Will a professional indemnity policy respond to damage caused by an agent your firm deployed? Put that to your broker in writing, before you need the answer. Policies written before agents were common may not address them either way. An ambiguous policy gets discovered at the worst possible moment.
Gather these documents before the legal review. Mark the clauses you want explained and identify the gaps you want negotiated. Reading the documents is preparation, not a substitute for advice about enforceability.
Which controls actually reduce the exposure?
Reduce the actions that can create avoidable harm, while treating liability as a separate legal question. Clear permissions, approval records and an incident process give your team concrete ways to control the workflow and explain what happened.
For example, require a person to approve a draft before any message can leave the system, and withhold sending access from the drafting process. Test whether another connection can bypass that boundary. A read-only connection limits changes through that route, but does not put a financial ceiling on every possible claim. Reading or disclosing information can itself cause harm.
Use these controls as practical review points:
- Approval gates on outbound and irreversible actions. Enforced in the system rather than requested in a prompt. What approval workflows do in an AI agent system covers the mechanism.
- Scoped access. The agent reaches the accounts and records the workflow needs and nothing else. What an AI agent can access covers how permissions are actually set.
- An audit log. When something goes wrong, you need to show what the agent did, when, and on whose authority. That is the difference between a contained incident and an unprovable mess.
For the data protection side of the same question, which runs on GDPR rather than liability law, GDPR for AI agents covers controller and processor roles. For what to do in the hours after a failure, what to do when your AI agent makes a mistake covers containment and recovery.
What this page cannot tell you
This page cannot tell you whether you are liable in a specific situation. That depends on your jurisdiction, your contracts, the facts of what happened, and in most of Europe on transposition legislation that is not final. Ask a qualified lawyer in your own country.
Keep the advice tied to your current jurisdiction, product and contract. Review it when the product changes, your business takes on a new legal role or the applicable national legislation changes. Assign someone to preserve the agreed terms and incident records, rather than relying on the agent’s own description after a failure.
What it can tell you is where to look, which questions to put to your provider and your insurer, and that the instrument most people assume protects them is written for someone else.
Frequently asked questions
Does the EU AI Act make my business pay compensation if my AI agent harms a customer? Article 85 creates a regulatory complaint route, not a civil compensation award. A customer’s damages claim needs its own applicable legal basis. Regulatory proceedings and a civil claim can arise from the same incident.[1]
Can my company sue an AI vendor under the EU Product Liability Directive? Not for the company’s own business loss under this directive. Article 5(2) separately permits claims through succession, subrogation or representation of an injured natural person. National law may also offer company remedies. Ask which route fits the claimant and loss.[2][3]
When does the new EU Product Liability Directive apply to AI systems? Member states must transpose it by December 9, 2026. Article 2(1) covers products placed on the market or put into service after that date. Check national commencement and transition rules; adoption of a bill alone is not proof that your product falls under the new regime.[2][6]
Is a business liable for what its AI agent does, or is the software vendor? It depends on the facts and legal role. The supplier may be a manufacturer; a business developing for its own use or substantially modifying a product can also enter that role. Contract, national liability and other applicable law remain relevant to the business’s own customer obligations.[2]
Notes
- Article 85: Right to lodge a complaint with a market surveillance authority. Regulation (EU) 2024/1689, June 13, 2024 Accessed October 15, 2026.
- Adopted Product Liability Directive, PE-CONS 7/1/24 REV1. October 23, 2024 Accessed October 15, 2026.
- Liability for defective products. Living guidance, October 2026 Accessed October 15, 2026.
- Procedure 2022/0303(COD). Withdrawal event October 6, 2025 Accessed October 15, 2026.
- Bundestag beschließt Modernisierung des Produkthaftungsrechts. October 8, 2026 Accessed October 15, 2026.
- Liability for defective products — implementation timetable. Living guidance, October 2026 Accessed October 15, 2026.
