AI Professional Indemnity Insurance: What It Covers

AI professional indemnity insurance pays for your defence and any compensation when a client says your AI work was negligent and cost them money, for example a model that gave wrong answers, a failed rollout or a biased system. It is ordinary professional indemnity cover, but the policy should name your AI work as insured activity so there is no argument later about whether it was meant to be covered.

What does professional indemnity insurance cover for AI work?

The Financial Conduct Authority describes professional indemnity insurance (PII) as cover designed for businesses that are required to compensate a party that claims to have suffered a loss as a consequence of negligent conduct or omission on the part of the business. For an AI developer or consultant, that means claims that your advice, design, code, model or implementation fell below the standard a competent professional would have met, and that your client lost money as a result.

Under section 13 of the Supply of Goods and Services Act 1982, which applies in England, Wales and Northern Ireland, where a supplier acts in the course of a business there is an implied term that it will carry out the service with reasonable care and skill. If a client says you broke that term, or owed them a duty of care in negligence and breached it, a PI policy is the cover that responds. A typical policy pays:

• legal costs of defending the claim, including expert evidence
• damages or an agreed settlement for the client’s financial loss

PI is about financial loss from your professional work. Injury to people and damage to property usually sit under public and product liability, and a data breach or hacking incident is usually a cyber matter. Our guide to what professional indemnity insurance is covers the basics, and professional indemnity vs public liability explains where the line falls.

Who needs AI professional indemnity insurance?

Any business whose AI work is relied on by a client to make decisions, serve customers or run part of its operation. In practice that includes:

• AI developers building models, agents, chatbots or integrations for clients
• AI consultants advising on strategy, tool selection, governance or rollout
• data scientists building forecasting, pricing, scoring or classification models
• digital and marketing agencies using AI to produce content, campaigns or customer journeys
• software firms adding AI features to products their customers depend on

If a client could reasonably say “we did what your system told us and it cost us”, you have a PI exposure. Many corporate and public sector clients also make PI, at a stated minimum limit, a condition of the contract.

What do AI claims against developers and consultants look like?

Most AI claims come from how a system was specified, tested, explained or handed over, not only from bad code.

• Model error. A demand forecasting model overstates sales, the client overorders stock and writes it off. The client says the model was trained on the wrong period and never properly validated.
• Hallucinated output. A customer service assistant you deployed gives customers incorrect refund terms or product information, and the client has to honour or compensate for what it said.
• Failed implementation. An automation project goes live, breaks an existing workflow and the client loses orders or pays staff to fix records by hand.
• Missed deadline. A delayed rollout means the client misses a launch date or a regulatory deadline and claims the lost revenue and extra costs.
• Bias. A screening or scoring tool treats one group less favourably and the client faces complaints or a claim, then turns to you.

Bias deserves extra care. The Information Commissioner’s Office says the Equality Act 2010 protects individuals from direct and indirect discrimination whether a decision is made by a human, an automated system or both. The ICO also says processing that leads to unjust discrimination will violate the data protection fairness principle, and that you should determine and document your approach to bias mitigation from the very beginning.

How do claims made policies and retroactive dates work?

PI is written on a claims made basis. The FCA explains that this means the current insurer is the one responsible for meeting claims made on the policy, even when the claims predate that policy. So the policy that responds is the one in force when the claim is made against you, not the one in force when you did the work.

• Keep cover continuous. A model you built three years ago can still produce a claim today, and if cover has lapsed nothing responds.
• Watch the retroactive date. The FCA describes a retroactive date as a date before which an insurer is unwilling to provide cover. If your retroactive date is last year, work you did before then is not covered, even under a current policy. When you switch insurers, ask to keep your existing retroactive date.

If you close the business or stop AI work, ask about run off cover so claims made after you stop still have a policy to respond.

What liability caps and indemnities do clients ask for in AI contracts?

AI contracts often contain three clauses that change your exposure, and your insurer will want to know about them.

• Liability cap. Most suppliers try to cap liability at a fixed sum or a multiple of fees. Under section 2(2) of the Unfair Contract Terms Act 1977, a business cannot exclude or restrict liability for negligence causing loss other than death or injury except so far as the term meets the requirement of reasonableness. Section 11(4) says that where a term restricts liability to a specified sum, the court looks at the resources you could expect to have available and how far it was open to you to cover yourself by insurance. Section 11(5) puts the burden of proving reasonableness on the party relying on the term. In plain terms, your level of PI cover can affect whether your cap holds.
• Unlimited liability for certain heads. Clients often ask for uncapped liability for data protection breaches, confidentiality and intellectual property. Section 2(1) of the same Act already prevents excluding liability for death or personal injury caused by negligence.
• Indemnities. Clients often want an indemnity against third party claims, for example if your model output infringes someone else’s rights or a regulator acts over the system you built. An indemnity can make you liable without the client proving negligence.

Check whether your policy covers liability you have accepted only because of a contract term, over and above what you would owe in law anyway.

How do you choose the right limit of indemnity?

The limit of indemnity is the most the insurer will pay. To choose one:

1. Start with the highest PI limit any current or likely client contract requires.
2. Look at the worst realistic loss your work could cause a single client: lost revenue while a system is down, remediation costs, regulatory costs and claims from their own customers.
3. Check whether the limit applies to each and every claim or in the aggregate for the year, and whether defence costs sit inside or on top of it.
4. Allow for several clients using the same model or product, since one flaw can produce several claims at once.

The price is driven by what you do with AI, how much of it is advice versus build, the sectors your clients work in, your turnover and fee income, the size of your largest contracts, your claims history, how you test, document and hand over work, where your clients are, and the limit and excess you choose.

What about clients and users in the United States?

US exposure is one of the first things an underwriter asks about. It can come from a US client, from a UK client whose customers are in the US, from software sold or downloaded there, or from a contract governed by US law. Insurers underwrite US exposure separately, so it needs to be disclosed, agreed and shown on the policy rather than assumed.

When you apply, say what share of fees comes from US clients, whether your AI output reaches US users, and which law your contracts choose. Cover for US exposure is available through Solar AI Cover.

Why should your policy name your AI work?

Many older PI and business policies were written before AI was common. They neither include nor exclude AI, which is often called silent AI. Silence can lead to a dispute after a claim about whether the insurer agreed to cover AI development, model training or automated advice when it priced the policy.

The safer approach is to describe your AI work fully on the proposal and have the policy name it as insured activity. Our guide does my business insurance cover AI explains how to check your current wording.

Example. A three person data science consultancy builds a churn prediction model that a subscription retailer uses to decide which customers get retention discounts. Six months later the retailer finds the model scored many loyal customers as high risk, says the training data was filtered wrongly, and claims the cost of the unneeded discounts and of rebuilding the model. The consultancy’s PI policy names “design, build and deployment of machine learning models for clients” as insured activity, the retroactive date goes back to when the firm started, and cover has been renewed without a break. The insurer appoints solicitors and an expert and the claim is settled within the limit. Had the policy said only “IT services”, the first weeks could have gone on arguing about scope.

What is Solar AI Cover and what do we need to quote?

Solar AI Cover is a scheme Solar Protect arranges with an insurer. The policy names your specific AI work as insured activity, rather than leaving AI unmentioned. Cover options include professional indemnity, cyber, product and public liability, and intellectual property infringement, and several of these exposures can sit in one policy. Cover for US exposure is available. All cover is subject to the insurer’s underwriting and the policy terms.

To quote we need:

• what your business does, including a clear description of the AI work
• your business address
• your turnover
• the limit of indemnity you need

Send us those details through our contact page and we will look for cover that fits your work and your contracts.

Frequently asked questions

Is AI professional indemnity different from normal professional indemnity?

The cover works the same way: it responds to claims that your professional work was negligent and caused a client financial loss. The difference is that the policy describes your AI work as insured activity, so the insurer has priced and agreed to it.

Does professional indemnity cover a data breach caused by my AI system?

Usually only in part. Breach response costs, notification and system restoration normally sit under cyber insurance. Our guide to cyber insurance explains the difference.

I use third party models such as large language models. Am I still liable?

Yes, to your client. If you chose, configured or deployed the model, the client will look to you if the output causes loss. Your rights against the model provider depend on its terms.

Do I need cover if my contract caps my liability?

Yes. A cap only works if it is reasonable under the Unfair Contract Terms Act 1977, and the cap still leaves you paying up to that amount plus your legal costs. Your PI limit should be at least as high as your cap.

What happens to my cover if I stop trading?

Because PI is claims made, a claim made after your policy ends will not be covered unless you buy run off cover. Ask about this before you close the business or stop AI work.

See also: AI Errors in Tax Returns: Who Is Liable?.

Sources

• FCA: Rma e professional indemnity insurance self
• Supply of Goods and Services Act 1982, section 13
• Unfair Contract Terms Act 1977, section 2
• Unfair Contract Terms Act 1977, section 11
• ICO: What about fairness bias and discrimination

Reviewed by Christine Anderson, Broker, Solar Protect. Last reviewed 31 October 2026. Solar Protect is the trading name of Solar Insurance Services (Medway) Limited, authorised and regulated by the Financial Conduct Authority. This page is general information, not advice on your own cover.