AI insurance

AI copyright and IP infringement insurance in the UK

Reviewed by the Apex broking team · Last reviewed 2026-10-26

AI IP infringement insurance usually covers defence costs and damages, subject to the policy terms, if someone claims your AI model, training data or output infringes their intellectual property. It matters because UK law on AI training and output is unsettled, and both model providers and the businesses using them can be sued.

Part of: AI insurance UK

In short

AI creates IP risk at three points: training data, the model and its output. In Getty Images v Stability AI (November 2025) the High Court held that a model whose weights do not store copies of training images is not an infringing copy, but did not decide whether UK training or copying outputs infringe, and the copyright point is under appeal. The government dropped its preferred training exception in March 2026 and proposes removing protection for computer-generated works under section 9(3) CDPA. Vendor indemnities are conditional, and IP infringement policies commonly exclude patents, deliberate infringement and contractual warranties.

Where the IP risk sits: training, the model and the output

AI creates intellectual property exposure at three points, and a different party may be sued at each. For the other AI exposures, see what AI insurance is.

Stage	What can go wrong	Illustrative claim
Training data	Copying works to train or fine-tune a model. The UK text and data analysis exception in section 29A CDPA is limited to non-commercial research.	A publisher alleges a start-up scraped its articles to fine-tune a commercial model
The model	An allegation that the model itself is an infringing copy, or that distributing it is a secondary infringement	A rights holder sues the developer for making model weights available for download in the UK
Output	Output reproduces a substantial part of a protected work, or carries someone's trade mark or logo	A marketing agency's AI-generated campaign image closely resembles a photographer's work

Open source code licences add another layer for coding tools: see open source licence liability. All scenarios above are illustrative.

Getty Images v Stability AI: what the High Court decided

In Getty Images v Stability AI [2025] EWHC 2863 (Ch), judgment given on 4 November 2025 by Mrs Justice Joanna Smith, Getty claimed that Stability AI's image model Stable Diffusion infringed its copyright, database right and trade marks.

Training and outputs were not decided. Getty accepted there was no evidence that training took place in the UK, and by closing submissions had dropped its primary copyright claims over training and outputs, and its database right claim.
Secondary infringement failed. Getty argued that importing and distributing the model in the UK was dealing in an "infringing copy". The court accepted that an "article" can be intangible, but held that the model weights do not store copies of Getty's works, so the model was not an infringing copy.
Trade marks: limited success. Getty won on limited, historic trade mark infringement where its watermarks appeared in outputs from certain older model versions. Its reputation-based claim failed, and passing off was not decided.

Getty was given permission to appeal on the secondary infringement point, which the judge described as a pure question of statutory construction; Stability was refused permission by the judge on the trade mark findings. Until the Court of Appeal rules, the case leaves open whether training in the UK, or outputs that copy a work, infringe. It does not make AI training "safe".

Who is liable if AI output infringes copyright?

Under section 16 CDPA, copyright is infringed by anyone who, without a licence, does a restricted act, such as copying, in relation to the whole or a "substantial part" of a work, or authorises another to do so. It does not matter that a model produced the copy. The government's March 2026 report on copyright and AI puts it plainly: "both the user and provider of a model may be liable for infringement of copyright".

In practice, the chain looks like this (illustrative):

the model developer, for training and for what the model outputs;
the business that deploys or builds on it, including through AI agents, for outputs it publishes or supplies;
the agency or consultancy, which may also have warranted to its client that the work is original and does not infringe;
the end client, who then looks to the agency under that warranty.

The contractual warranty is often the bigger exposure, because it can make you liable without proof of copying. See AI-generated content and IP infringement and how PI treats contractual liability.

Who owns AI output: CDPA section 9(3)

Section 9(3) CDPA says that for a literary, dramatic, musical or artistic work which is computer-generated, the author is "the person by whom the arrangements necessary for the creation of the work are undertaken". Section 178 defines computer-generated as generated by computer where there is no human author, and section 12(7) gives such works 50 years' protection from the end of the year they were made.

The provision is still in force, but its future is uncertain: the government's March 2026 report proposes removing this protection for computer-generated works, while keeping protection for AI-assisted works with a human author. If you promise clients they will own what your AI produces, check that the promise matches the law and your contracts.

UK government policy on AI and copyright (October 2026)

The government consulted on copyright and AI from 17 December 2024 to 25 February 2025, including a proposed exception for text and data mining with a rights holder opt-out.
In its Report on Copyright and Artificial Intelligence, laid before Parliament in March 2026 under the Data (Use and Access) Act 2025, the government said a broad exception with opt-out "is no longer the government's preferred way forward". No new AI training exception has been proposed.
It favours greater transparency about training material, working with industry on good practice rather than immediate legislation.
Legislative change remains possible, but no timetable has been set.

Firms placing general-purpose AI models on the EU market should also note that the EU AI Act has required providers of those models, since 2 August 2025, to have a copyright policy and publish a summary of training content.

Vendor IP indemnities and their limits

Some AI vendors offer business customers an IP indemnity, promising to defend them against third-party copyright claims over outputs. These commitments help, but they are usually conditional, for example on the customer having switched on the vendor's content filters and other required safeguards, and on the customer being able to show it did so if a claim arises.

Common limits to look for in any vendor indemnity:

it is limited to the vendor's own model and output, not to your fine-tuning data, prompts or added content;
it depends on you following the vendor's guardrails and usage terms, and you may have to prove it;
it may be capped, or excluded where you knew or should have known the output infringed;
it does nothing for the warranties you give your own clients.

An indemnity is a contract with the vendor, not insurance. Your own cover sits behind it.

How IP infringement cover usually works

IP infringement cover, written standalone or as an extension to professional indemnity, commonly pays defence costs and damages or settlements for claims alleging unintentional infringement of copyright, trade marks, design rights or database rights, and sometimes passing off. See the IP infringement cover definition. Common exclusions include:

Patents, often excluded or tightly limited;
deliberate or knowing infringement, including continuing after a complaint;
contractual liability beyond what you would owe at law, such as broad IP warranties and indemnities to clients;
licence fees and royalties you would have had to pay anyway;
claims known before the policy started, and in some wordings, claims brought in the USA or Canada.

Insurers are also adding AI-specific wording. Check whether your policy is silent or affirmative on AI, and whether training data and generated output are within the insured activities. For start-ups, see insurance for AI start-ups.

How Apex's AI insurance scheme covers IP infringement

Apex has an AI insurance scheme with an insurer in which your specific AI work, such as training models, building on third-party models or producing AI-generated content for clients, is written into the policy as insured activity rather than left to a generic business description. Cover options are professional indemnity, cyber, product / public liability and IP infringement, so you pick the options you need, 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, or read the AI insurance guide.

Sources

Frequently asked

Who is liable if AI output infringes copyright in the UK?

Potentially both the provider of the model and the business that uses or publishes the output. The UK government's March 2026 report says "both the user and provider of a model may be liable for infringement of copyright". If you supplied the output to a client with a warranty that it does not infringe, the client can also claim against you under the contract.

Is it legal to train AI on copyrighted material in the UK?

There is no general exception for commercial AI training. Section 29A CDPA allows text and data analysis for non-commercial research by someone with lawful access, not for commercial use. The government dropped its preferred opt-out exception in March 2026 and has not proposed a replacement. Getty v Stability AI did not decide the point because the training happened outside the UK.

What did Getty Images v Stability AI decide?

In November 2025 the High Court rejected Getty's secondary copyright claim, holding that Stable Diffusion's model weights do not store copies of Getty's images and so are not an infringing copy. Getty won limited trade mark claims over watermarks in outputs from older model versions. Getty has permission to appeal the copyright point.

Who owns copyright in AI-generated content in the UK?

Where a work is computer-generated with no human author, section 9(3) CDPA treats the author as the person who made the arrangements necessary for its creation, with 50 years' protection. Where a human uses AI as a tool, normal authorship rules apply. The government has proposed removing protection for purely computer-generated works.

Does professional indemnity insurance cover AI copyright claims?

Some PI policies include IP infringement cover for unintentional infringement in your professional work, but many limit or exclude it, and some now carry AI exclusions. Patents, deliberate infringement and contractual warranties are commonly excluded. Check that your AI work is named in the policy and that the IP cover extends to AI-generated output.

Get a quote for AI insurance

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 callback

Apex 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.