FCA authorised · FRN 724952 0117 325 0027Quote & buy →
AI insurance
AI agent insurance is cover for the liability and losses that arise when AI software acts on its own, such as sending messages, making payments, booking services or changing code. It matters because the early cases treat an AI system's statements as the business's own, and an agent that acts can cause loss faster and at greater scale than one that just advises.
Part of: AI insurance UK
In short
An AI agent does not carry liability itself; the businesses around it do. In Moffatt v Air Canada (2024) a Canadian tribunal rejected the idea that a chatbot was responsible for its own statements and held the airline liable for negligent misrepresentation. Loss from an agent's action usually lands on the deployer, then moves along contracts to the vendor or integrator if the contracts allow. Which insurance responds depends on the loss: professional indemnity for clients' financial loss, cyber for data and system incidents, product or public liability for injury and damage, and IP cover for infringement.
A chatbot answers questions. An AI agent is given a goal, tools and permissions and acts: it can send emails, update records, make or approve payments, book services, run code or change a live system. The ICO's Tech Futures report on agentic AI, published in January 2026, describes agentic AI as combining generative AI "with additional tools and new ways of interacting with the world", and says organisations "remain responsible for data protection compliance of the agentic AI they develop, deploy or integrate".
Two things change the liability picture:
On 14 February 2024 the British Columbia Civil Resolution Tribunal decided a claim over an airline chatbot that told a customer he could claim a bereavement fare within 90 days of buying his ticket, after travelling; the airline's own policy page said otherwise. The airline argued it could not be liable for its chatbot's information, which the tribunal said in effect suggested "the chatbot is a separate legal entity that is responsible for its own actions". The tribunal rejected this: "It should be obvious to Air Canada that it is responsible for all the information on its website" (para 27). It found the airline "did not take reasonable care to ensure its chatbot was accurate" (para 28), held it liable for negligent misrepresentation and awarded C$650.88 in damages, C$812.02 in total with interest and fees.
Singapore's Court of Appeal held that a cryptocurrency exchange was not entitled to reverse trades made automatically by trading software. For contracts made by deterministic algorithms, knowledge of a mistake is assessed "by reference to the state of mind of the programmers of the algorithms at the time of the programming" (para 97). Agents built on large language models do not always behave deterministically, so how that test applies to them is open.
On 12 July 2024 a US federal court in California refused to dismiss claims that an AI hiring-screening vendor could be liable as an agent of the employers using its tools, finding it plausible that employers had "delegated their traditional function of rejecting candidates" to the software. A collective age discrimination claim was allowed to proceed in May 2025 and a June 2026 ruling let further claims continue. The rulings are pre-trial, but show vendors can face direct claims as well as contractual ones.
None of these decisions binds UK courts. For consumer services, section 50 of the Consumer Rights Act 2015 treats anything "said or written to the consumer, by or on behalf of the trader" about the trader or service as a contract term if the consumer takes it into account, and section 57 stops the trader excluding that liability. An agent speaking for a business is arguably speaking on its behalf. Between businesses, negligent misstatement can produce a similar result. The Law Commission's discussion paper Artificial Intelligence and the Law (31 July 2025) treats AI autonomy as raising questions of causation and liability, and says it is not clear any current AI system is advanced enough to warrant legal personality. It makes no reform proposals.
| Party | Role | How the loss reaches them | Cover that may respond |
|---|---|---|---|
| Deployer (the business using the agent) | Chooses the agent, sets its permissions, presents it to customers | Directly liable to customers and third parties for what the agent says and does, as in Moffatt | PI for clients' financial loss; cyber for data and system incidents; product or public liability for injury or damage |
| Vendor or model provider | Builds or supplies the agent or model | Through contract with the deployer, usually capped; possibly directly, as alleged in Mobley | Technology PI, cyber, IP infringement; product liability where software is part of a physical product |
| Integrator or AI consultant | Configures, connects and tests the agent | Claims for negligent design, configuration or advice | PI |
| Client or end user | Instructs or relies on the agent | Bears its own loss where the contract or its own fault puts it there | Its own policies, if any |
In practice the customer pursues the deployer, because it contracted with it and dealt with its agent. Whether the deployer can pass the loss up the chain depends on the vendor contract.
Most agent losses are settled by contract terms, not AI law. Check vendor and client contracts for:
Be careful what you promise. PI policies often exclude liability taken on by contract beyond what you would owe anyway. See how PI treats contractual liability and insurance requirements in technology contracts.
| Agent incident (illustrative) | Main loss | Cover most likely to respond, subject to the policy terms |
|---|---|---|
| Emails a client's confidential file to the wrong recipient | Breach response costs, regulator inquiry, client claim | Cyber; PI if the error is in professional services to that client |
| Pushes a code change that takes a client's system offline | Client's lost revenue and recovery costs | PI (technology errors and omissions) |
| Promises a customer a refund the business does not offer | Cost of honouring it or compensating the customer | Often an uninsured business cost; a third-party misrepresentation claim may fall under PI, depending on the wording |
| Controls a robot or connected device that injures someone | Bodily injury or property damage | Product or public liability. See AI product and public liability |
| Publishes images or copy that reproduce someone else's work | Copyright or trade mark claim | IP infringement cover. See AI IP infringement insurance |
Check every policy for AI exclusions; see affirmative AI cover vs silent AI and does PI cover AI mistakes.
Under section 3 of the Insurance Act 2015 you must disclose every material circumstance you know or ought to know before cover starts. For agents, expect questions on:
Keep records that support your answers. See what to document about your AI use and the duty of fair presentation.
Apex has an AI insurance scheme with an insurer that writes your specific AI work, such as the agents you build or deploy, into the policy as insured activity rather than leaving it to a generic business description. You choose from professional indemnity, cyber, product and public liability, and IP infringement cover, so several agent exposures can sit in one policy, and cover for US exposure is available, subject to the insurer's underwriting and the policy terms. To quote, we need a precise description of your AI work, your address, your turnover (including how much comes from the USA and Canada) and the limit of indemnity you want; you can start an AI insurance quote online in a few minutes. See AI insurance UK, what is AI insurance and US cover for UK AI firms.
See also: AI chatbot insurance · Agentic AI liability
Usually the business that deployed the agent is the one its customers pursue, because the agent acts on its behalf. In Moffatt v Air Canada (2024) a tribunal held the airline responsible for its chatbot's wrong answer. The deployer may then recover from the vendor or integrator, as far as their contracts and liability caps allow.
It is unlikely to work. In Moffatt v Air Canada the airline argued it was not liable for its chatbot; the tribunal rejected this and said the airline was responsible for all the information on its website. In the UK, section 50 of the Consumer Rights Act 2015 can make statements made on a trader's behalf binding contract terms.
It can, if the agent's work falls within your insured professional activity and no AI or other exclusion applies. PI responds to clients' financial loss caused by your error, subject to the policy terms. It does not usually cover your own losses, such as money an agent pays away, or injury and property damage.
Sometimes, but usually within the limits its contract sets. Vendor terms commonly cap liability and say outputs must be checked. Direct claims are also possible: in Mobley v Workday a US court allowed claims that an AI screening vendor acted as the employers' agent to proceed in 2024.
Yes, if it is material, and agents that act without human approval usually will be. The Insurance Act 2015 requires a fair presentation of the risk. Describe what the agents do, their permissions and controls. Leaving out material facts can lead to a reduced claim payment or no cover.
Tell us what your AI work is, where you are based, your turnover and the limit you need. It takes a few minutes online, and a broker will come back to you with options. Or call 0117 325 0027.
Start an AI insurance quote Request a callbackApex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information, not advice on your individual circumstances, and it does not guarantee that cover will be available or on what terms.
