AI-related disputes

When an AI system causes loss, there is no AI liability regime to turn to. There is contract, tort and data protection — and they answer more than people expect.

What these disputes look like

The output was wrong and someone relied on it. A model produced a figure, a citation, a translation or an analysis that was false, and a decision was taken on it.

The system did not do what was sold. Accuracy claimed in procurement is not accuracy delivered in production — often because the vendor’s benchmark data looked nothing like yours.

The model degraded. It worked at deployment and performs worse now, because the underlying model changed, or the world did.

The system discriminated. A hiring, credit or access decision fell disproportionately on a protected group. This is usually an employment or Equality Act claim rather than a technology one — see the discrimination pages.

The data should not have gone in. Confidential or personal data entered a system that retained it, trained on it, or exposed it.

The questions a court will ask

What was actually promised? Marketing language rarely survives an entire agreement clause. What matters is the warranties, the specification, and any service levels tied to performance.

What was excluded, and does the exclusion hold? Vendor terms usually exclude accuracy, consequential loss and fitness for purpose. Whether those exclusions bind depends on incorporation, and on reasonableness where the Unfair Contract Terms Act applies. Between commercial parties the bar is high — but not unreachable, particularly where the exclusion would leave the buyer with no remedy at all for the very thing they bought.

Was the reliance reasonable? This is where most of these disputes turn. If a system’s documentation says outputs require verification and the buyer did not verify, the loss is substantially their own. If the system was sold as removing the need for review, that is a different case. The output warning in the documentation is the vendor’s strongest defence and the first thing we read.

Who owed a duty, and to whom? Where loss falls on a third party — a customer, a candidate, a patient — the chain runs through the deployer first. The organisation that used the system is the one the claim lands on, and whether it can pass that on is a separate question decided by the contract.

Why the law is thinner than the coverage suggests

There is very little decided English authority on AI performance. That has two practical consequences.

Outcomes are less predictable, which raises the value of early, realistic assessment and often makes settlement the right answer earlier than in a conventional contract dispute.

Evidence matters more than usual. What the system was told to do, what it produced, what the logs show, what the vendor said in procurement, what was disclosed at the time — these decide the case. Preserve them immediately. Model versions change and logs are often kept for weeks, not years. This is the single most consequential thing a party can do in the first fortnight.

What we do

Assess the position, preserve the evidence, and advise on whether the claim is worth bringing or defending — with the same directness we apply to any commercial dispute. Where the exposure is discrimination rather than contract, the employment and Equality Act teams lead.

Questions

Questions people ask us

Can I sue an AI vendor for a wrong output?

Sometimes. It depends on what was warranted, what was excluded, and whether your reliance was reasonable. A vendor who sold verified accuracy is in a very different position from one whose documentation told you to check.

The contract excludes liability for accuracy. Is that the end of it?

Not necessarily. Exclusions must be properly incorporated and, where UCTA applies, reasonable. An exclusion that removes any remedy for the core thing purchased is more vulnerable than a standard consequential loss cap.

Our staff relied on a hallucinated output and we lost money. Whose fault is that?

Partly yours, almost certainly — the question is how much. If the tool was sold as needing no review, less. If its own documentation said to verify, more.

What should I do first?

Preserve everything: prompts, outputs, logs, model version, procurement correspondence. Do it now. This evidence disappears quickly and cases are decided on it.

The hub

AI law →

Also in this area

AI in recruitment and automated decisions →

Also in this area

Does the EU AI Act apply to your UK business? →

Start with a conversation

A free 20-minute call. Tell us what has happened and we will tell you whether we can help, what it would involve and roughly what it would cost.

No charge

A free 20-minute call

Tell us what has happened and we will tell you whether we can help, what it would involve and roughly what it would cost. No advice is given on this call and there is no charge for it.

£350 plus VAT

A paid strategy session

One hour with a partner, followed by a written summary of your position and options. For people who want proper advice without instructing a firm yet. Credited in full against your fees if you go on to instruct us.

Or reach us directly

0207 112 8818

WhatsApp +44 7476 045967

[email protected]

We answer enquiries the same working day.