AI product liability terms for a startup

A start-up selling an AI product sells something that will sometimes be wrong, cannot promise otherwise, and will be judged by customers, regulators and investors on how its terms handle that. The terms have to describe the product accurately, be honest about accuracy without disclaiming everything, respect the different rules for consumers and business customers, restrict the uses that could cause harm, set a cap and an indemnity structure an investor will accept, and anticipate the regulation that follows the product into the markets it sells to. The agreement's AI provisions come back drafted or reviewed within five working days for a fixed £595.

Who this is for

Early-stage and growth companies in England and Wales selling AI products or AI-enabled services to consumers or businesses, and founders whose investors have asked to see the customer terms.

What matters in AI product liability terms

What the product is sold as and the warranty that follows

The terms describe the product as what it is (a tool that generates, assists, recommends, predicts or automates, within a stated scope, for stated users, with a person expected to review where the terms say so) and warrant that: that the product will perform materially as described with reasonable care and skill in its operation, that the company has the rights to provide it, and that it complies with the stated security and data protection commitments; the warranty does not extend to the accuracy of particular outputs, to fitness for a purpose the customer did not disclose, or to results obtained from outputs the customer altered; a product marketed as 'accurate' has made a warranty the terms cannot take back.

The accuracy problem and the clause that is honest about it

The terms state that outputs are generated by statistical models, may be inaccurate, incomplete, biased or inappropriate, should be reviewed by a competent person before reliance, and are not a substitute for professional advice where the product's domain is one in which advice is regulated; they state the measures the company takes (testing, evaluation, monitoring, the ability to report problems) and that the company will correct material defects in the product as the warranty provides; the clause is drafted to be true of the product rather than a boilerplate disclaimer, because a disclaimer that contradicts the product's marketing or its interface (a confident answer with no warning) is unlikely to be enforced against a consumer and may fail the reasonableness test for a business customer.

Consumers, business customers and the two sets of rules

For consumers the Consumer Rights Act 2015 requires digital content to be of satisfactory quality, fit for purpose and as described (sections 34 to 36) and services to be performed with reasonable care and skill (section 49), prohibits excluding those rights, treats unfair terms as not binding and gives the consumer repair, replacement, price reduction and refund remedies, so the terms limit liability only to the extent the Act permits and exclude only what can lawfully be excluded; for business customers the Unfair Contract Terms Act 1977 applies the reasonableness test to exclusions in standard terms, so the terms cap liability at a level connected to the fees and the insurance, exclude indirect loss, and define the customer's responsibilities; one set of terms for both audiences usually fails one of them.

Decisions, harm and the uses the terms must restrict

where an AI tool alone makes a decision with legal or similarly significant effects on a person, Article 22 of the UK GDPR as amended by the Data (Use and Access) Act 2025 applies, allowing it outside special category data only with the safeguards of information, representations, human intervention and contest, and the terms prohibit customers from using the product for such decisions without the safeguards, prohibit safety-critical, medical, legal and financial uses the product is not designed for, require human review where the product's domain carries risk of harm, and state the uses the product supports; a product that could cause physical harm (one that controls equipment, advises on medication, assesses safety) needs its own analysis, because the Consumer Protection Act 1987 imposes strict liability for defective products that cause injury or damage and the common law of negligence applies to the company's design decisions regardless of the terms.

Insurance, indemnities and the cap an investor will read

The company's liability is capped at a stated multiple of the fees paid (or a fixed sum for free tiers), with the cap connected to the professional indemnity, cyber and product liability insurance the company holds (which may exclude AI-specific risks unless the policy is chosen for them), indirect and consequential loss excluded, the customer indemnifying the company for its inputs and its use, and the company's indemnity (if any) limited to third-party intellectual property claims about the product itself, not about outputs; the terms are drafted with a position the company can hold and a negotiating range the founders understand; the author of a computer-generated work is, under section 9(3) of the Copyright, Designs and Patents Act 1988, the person who made the arrangements, a rule untested on modern tools, and the Act protects only what is original, so the terms also tell the customer what it does and does not get in outputs.

Regulation, the EU and the obligations that follow the product

The United Kingdom regulates AI through existing regulators (the Information Commissioner, the Competition and Markets Authority, the sector regulators) applying existing law, while the European Union's AI regulation imposes obligations on providers of AI systems placed on the EU market, with prohibited practices, high-risk categories carrying conformity obligations, and transparency duties for certain systems, so a start-up selling into the EU needs to classify its product and build the obligations into its terms and its product; the terms state the markets the product is offered in, require the customer to comply with the regulation applicable to its own use, and reserve the company's right to restrict features by market; the AI supplier review service covers the model providers the product depends on.

What it costs

AI provisions in a contract, £595. Review or draft AI-specific terms in a supply or customer agreement. Five working days.

AI use policy, £995. Internal policy plus the accompanying contract provisions. Five working days.

Email me at geoffrey@caesar.co.uk with what you need and any documents, and I will confirm the scope and the fee in writing, usually the same working day.

What you get

• An internal AI use policy your staff will follow, rather than a document that bans everything and is quietly ignored
• Contract provisions covering output ownership, training data, confidentiality and liability
• A defensible position to give customers who ask what your AI governance looks like
• Review of your AI suppliers' terms, including what they reserve the right to change
• Practical guidance on what to permit, what to restrict and what to prohibit outright
• One round of amendments

What is not included

• Technical AI assurance, model testing or bias auditing
• Sector-specific regulatory approval
• Advice on non-UK regulatory regimes beyond flagging where they apply
• Ongoing regulatory monitoring

Questions I am often asked

Can we disclaim liability for our product's outputs being wrong?

Not entirely: for consumers the Consumer Rights Act prevents it, and for business customers a total exclusion fails the reasonableness test. The clause is honest about accuracy, warrants the product as described and caps liability at a level connected to fees and insurance.

What cap will investors and enterprise customers expect?

A stated multiple of fees connected to your insurance, indirect loss excluded, an indemnity limited to IP claims about the product rather than outputs. Enterprise customers push for more, and the terms are drafted with a position you can hold and a range you understand.

Does the EU AI regulation apply to us?

If your product is placed on the EU market, yes, with obligations by category. The terms state your markets, require customers to comply for their own use and let you restrict features by market; the classification itself needs its own analysis.

Related guidance and services

• AI contracts and AI use policies, £595, the service this page describes
• SaaS and technology contracts, £995
• Data protection agreements and privacy terms, £795
• AI features addendum to SaaS terms
• AI development agreement for a bespoke model

This page is general guidance for businesses in England and Wales, not advice on your own circumstances. Last reviewed: October 2026. Email geoffrey@caesar.co.uk.