Issued against a backdrop of rapidly increasing AI adoption and limited judicial authority on the subject, the UK Jurisdiction Taskforce (UKJT) has published its long-awaited final legal statement on liability for AI harms following a public consultation earlier this year.
The overall theme of the final statement is that the established principles of English Law are sufficient to address the potential harms of AI, without the need to develop a specific legal framework.
The UKJT defines AI as ‘a technology that is autonomous’ but which does not have legal personality in English law and therefore cannot be held liable in its own right.
The statement sets out two sources of liability for AI-caused loss:
- Voluntarily assumed responsibilities i.e. a contract
- Liability imposed by law regardless of whether a party has chosen to take on that liability
In the absence of a contractual framework, negligence is likely to remain the principal route by which AI related harms are evaluated.
What’s changed between the draft and final statement?
Although the UKJT’s core conclusions remained largely unchanged following consultation, the final legal statement is significantly more detailed. One of the most important additions is an entirely new discussion of non-delegable duties – a topic not addressed at all in the consultation draft, which covered only vicarious liability. The final statement highlights that organisations in sectors such as healthcare may remain liable for AI-related harm even where the relevant technology has been supplied by a third party. The UKJT cites the example of an NHS Trust that cannot escape liability for an AI diagnostic tool's defect by pointing to the fault of the developer. Helpfully, the statement also concludes that a party found liable under a non-delegable duty may seek a contribution from the party actually at fault (such as the AI developer) under the Civil Liability (Contribution) Act 1978. The final statement also substantially expands its analysis of causation, recognising that these questions are likely to be among the most challenging and heavily litigated issues as AI-related disputes emerge.
In addition, the UKJT strengthens its discussion of professional responsibility, placing greater emphasis on due diligence, validation of outputs, and compliance with ever evolving professional standards.
Professional negligence in an AI-enabled world
Of particular interest to professionals is the UKJT's position on the negligent use of AI.
The statement confirms that a professional's obligation to exercise reasonable skill and care extends to the use of AI just as it applies to any other professional tool. The UKJT identifies a number of situations in which professional negligence might arise, including failures to:
- Undertake adequate due diligence on an AI system
- Understand the limitations of the technology
- Be transparent with clients as to the use of AI
- Protect confidential or privileged information which is entered into AI systems
- Ensure there has been sufficient testing of an AI system to check it is suitable and appropriate for the task envisaged
- Maintain appropriate human oversight
AI does not lower the professional standard and interestingly there are situations where a professional could be held liable for failing to use AI for a task, where a competent member of their profession would have done so. The UKJT gives several of its own examples: a radiologist who fails to use an AI system that is highly effective at identifying cancerous tumours and could have been procured at reasonable cost; an auditor who fails to use AI to detect anomalies and fraud across a very large volume of similar transactions; and a solicitor in the Business and Property Courts who fails to advise their client to consider an AI-assisted tool to review large volumes of documents. This issue is only going to become more relevant as AI tools become more efficient and commonplace.
The difficult question of causation
The UKJT states that the general rules of causation, including the traditional "but for" test, can in principle be applied to AI harms, and that in many cases the autonomous nature of the tool has no bearing on what caused the loss suffered. However, the UKJT is also clear that AI's autonomy and opacity can genuinely make it difficult to establish why a particular outcome occurred as parties seek to establish the cause of a loss – a difficulty it regards as no more severe than in non-AI cases.
The UKJT envisage expert evidence helping to bridge the gap here through experimentation, i.e. AI prompts can be reviewed and different scenarios can be generated to test how the AI arrived at the output that was generated.
Of course, there will be situations where it will not be possible for a party to prove how a loss was suffered, and the UKJT is clear that the law is capable of evolving and/or developing exceptions to meet these challenges, as it has done in the past. The statement points to two distinct lines of authority. First, in Fairchild, where it was scientifically impossible to identify which of several exposures caused the claimant's mesothelioma, the House of Lords held that it was enough to show that a defendant's breach had materially increased the risk of injury. Second, and separately, English law also recognises a "material contribution to damage" doctrine, recently endorsed more generally by the Supreme Court, which applies where multiple wrongdoers each contribute to a harm that would have occurred in any event, so no single contribution can be shown to be a "but for" cause. The UKJT's view is that there is no reason why either approach couldn't be applied to harms caused by AI where the “but for” test is not practical.
At the same time, the partial defence of contributory negligence should still be available in the context of AI harm, although non-commercial users of AI are less likely to be found to be negligent.
It is also worth noting what the final statement says professionals should not expect: strict, "no fault" liability under the Consumer Protection Act 1987 is unlikely to apply to most professional AI tools. That regime is confined to defective tangible products, and the UKJT's view is that pure software and AI supplied as a service – including most chatbots and generative AI tools professionals actually use – will fall outside it because they are unlikely to be treated as “goods”. For professionals, fault-based negligence therefore remains the dominant liability framework, reinforcing the importance of the due diligence and oversight standards discussed above.
Looking ahead
The UKJT's statement is not legally binding. Indeed, the UKJT itself is explicit that the statement "is not intended to be legal advice" and that "nothing in it should be relied upon as being relevant to any particular circumstances". Nonetheless, it is likely to be highly influential and there is every reason to expect that it will help to shape judicial thinking on AI related disputes.
For professionals embracing AI, that means the focus should remain on familiar concepts: competence, oversight, and risk management. The technology may be new, but the legal principles are not.
The full text of the final legal statement is available here.
This article is a part of our FutureProof series. If you're interested in finding out more about how AI is impacting professional life, you can follow that here.
Our content explained
Every piece of content we create is correct on the date it’s published but please don’t rely on it as legal advice. If you’d like to speak to us about your own legal requirements, please contact one of our expert lawyers.