In brief
- The UK Jurisdiction Taskforce (UKJT) has now published a final version of its legal statement on liability for AI harms, following a public consultation earlier this year.
- The UKJT's position remains largely unchanged from its draft statement: in most cases liability for AI harms will be governed by existing principles of contract or tort law. The UKJT expects that English law is sufficiently flexible to accommodate any novel issues arising from AI harms.
- However, the final statement does expand on the analysis in some key ways, including: the interplay between AI harms and non-delegable duties of care; the application of the "material contribution to damage" approach to causation; and the availability of contributory negligence as a partial defence to claims arising from AI harms.
- It also reframes the approach taken in the draft statement to liability for defamatory statements published by AI.
Introduction
We previously wrote about the UKJT's draft legal statement on how English law can and should approach novel questions of liability arising from the use of AI tools.
Following a consultation process, the UKJT has now published its final legal statement. Broadly speaking, the UKJT has maintained its earlier positions. For example, it continues to adopt a "technology agnostic" definition of AI: "technology that is autonomous". It also remains of the view that, as "a well-developed flexible common law system", the English legal system is able to "provide certainty and predictability in the context of technological innovation".
As before, the UKJT concludes that questions of liability resulting from the use of AI will primarily be dealt with via either contract law, or the tort of negligence. Where harm arises that is subject to contractual agreements, the final statement emphasises that "the fact that the subject matter of a contract may be AI poses no special difficulty". Similarly, with regard to the law of negligence (and other tortious claims), it concludes that:
"There is no conceptual reason why the law of negligence cannot be applied to harms caused by AI failures to ascertain whether and to what extent a party will be liable for those harms. In many cases, neither is there any practical difficulty in applying the normal principles in an AI context".
The UKJT's overall approach, and much of its detailed analysis, remains unchanged in its final statement. However, some further elements have been added which develop their analysis.
Key changes
The final statement adds analysis of the following points:
- Liability for AI harms in the context of "non-delegable duties";
- The "material contribution" approach to causation;
- The application of contributory negligence; and
- A "participation or authorisation" approach to liability for the publication of defamatory material.
Vicarious liability and non-delegable duties
In its draft statement the UKJT concluded that an AI has no legal personhood. Therefore, AI cannot itself be held liable for harms caused by its use.
By extension, the UKJT concluded, there cannot be vicarious liability for the actions of an AI tool: a party can only be vicariously liable for the actions of another legal person that is itself liable for the harms caused by those actions. An AI or AI tool is not a person, and so such vicarious liability cannot arise. There could, however, be vicarious liability for another person's negligent misuse of AI.
The final statement maintains this analysis. However, the UKJT now considers the situation where a person owes a "non-delegable" duty to protect a third party from harm. As examples of such duties, it cites the duty of an employer to "provide a safe system of work for its employees", or for hospitals to safely treat their patients. In these circumstances, the UKJT concludes, the party that owes such a duty is likely to be liable where an AI has caused harms that it was duty bound to protect against. This is so even if they have not been negligent in allowing the use of AI.
Material contribution
The final statement also expands the analysis of the application of the law of causation to harms caused by AI. As with any cause of action, where a claimant seeks damages for harms arising from the use of AI it must show that the development of faulty AI and/or the (mis)use of AI has caused the harm suffered. The UKJT had acknowledged in the draft statement that the "opaque" way in which AI operates might make causation particularly challenging to prove. However, it concluded that these challenges "are neither more severe nor different in kind to the sorts of issues that arise in other domains and that the English common law is well able to accommodate".
Expanding on this line of analysis, the final statement now also highlights the English law principle of "material contribution to damage". This applies where: multiple parties have each, in breach of a duty of care, contributed to a single harm; but, no individual contribution was on its own necessary or sufficient to cause that harm. In these limited circumstances, English law does not insist on proving a strict "but for" causation, which would require the claimant to prove that the harm would not have occurred had that particular breach not occurred. Instead, it may be enough to show, on the balance of probabilities, that a party's actions "materially contributed to the damage".
The UKJT points out that: this material contribution principle already applies in some contexts, such as industrial disease and clinical negligence claims; and, the Supreme Court has recently endorsed it as a more general principle. The UKJT states there is "no reason in principle" why this approach could not be adopted in relation to harms caused by AI. This would potentially make it easier to establish liability even in the case of harms which have or might have multiple causes.
Contributory negligence
The draft statement did not consider the application of contributory negligence in the context of harms caused by AI.
The final statement now includes a section specifically addressing this principle. It concludes that contributory negligence will be available as a partial defence in this context in the same way as any other. The final statement gives the example of a party that uses AI to plan a driving route, and then suffers harm as a result of following the proposed route. If the user followed the AI's route despite knowing that it would expose them to significant hazards that could be avoided by another route, then it is likely that any damages awarded would be reduced as a result of their own contributory negligence.
Liability for publication
Perhaps the most significant change in the UKJT's analysis is in relation to liability for publication of defamatory material produced by AI.
In its draft statement the UKJT framed its analysis in terms of "primary" and "secondary" publishers. "Primary" publishers are those that create the defamatory material, and "secondary" publishers are those that disseminate defamatory material created by others. The UKJT queried whether the creator of an autonomous AI could be considered a "primary" publisher of material generated autonomously by the AI, or if they were more likely to be treated as "secondary" publishers, which are "protected by statute".
In the final statement the UKJT has adopted a different approach. Instead of distinguishing between "primary" and "secondary" publishers, it suggests that a person will be considered a publisher if they have "participated in, or authorised" the publication. It notes that the question of whether a party involved in the "AI supply chain" will meet this threshold will be "highly fact-specific". The UKJT suggests that a party will be more likely to do so (and therefore potentially attract liability) if it:
- Manually reviews the AI's output before publication; or
- Deliberately deploys the AI for the purpose of publishing statements to the public in the course of its business.
However, it notes that there will be a number of factual and legal considerations that will have to be applied to the specific facts of each case.
Conclusion
The UKJT's final statement remains very close to the draft statement published for consultation. It expands the analysis in some areas, and in the case of the discussion of defamation, significantly reframes the issues. However, overall, the conclusion remains clear: English law is flexible. AI presents some novel challenges, and provides a new context to existing ones. In either case, the authors remain confident that the law will be able to adapt to and accommodate claims relating to AI based harms.
The law in this area is already beginning to evolve: as the UKJT also notes, the EU has published and dropped an AI Liability Directive. Meanwhile some professional bodies (such as the Royal Institution of Chartered Surveyors) have already published guidance on appropriate use of AI systems which will likely frame the courts' assessment of liability in these areas. Professionals in particular should also take heed of the UKJT's warning that they "could be liable for failing to use AI in circumstances where a competent member of their profession would have done so".
It still remains to be seen how the English law will react when claims arising from AI harms reach the courts. However, whilst not binding, the UKJT's statement is likely to be considered by the courts when they do come to grapple with these issues. No doubt it will be persuasive.