UK Jurisdiction Taskforce publishes final legal statement on liability for AI harms

The UK Jurisdiction Taskforce (UKJT) has published its final legal statement on liability for AI harms under the private law of England and Wales, following a public consultation earlier this year. The statement seeks to address in what circumstances, and on what legal bases, English law will impose liability for loss that results from the use of AI.

The UKJT takes the view that English law is flexible, has frequently accommodated novel and disruptive technological developments, and has shown that it can provide certainty and predictability in the face of technical innovation. 

The overall message of the statement is that English law is generally well-equipped to respond to the types of liability questions that AI may in due course present, although the application of existing legal concepts to particular issues will always be highly fact-dependent and some areas of uncertainty are acknowledged.

The full text of the statement is available for download here. The statement is not legally binding, nor is it intended to be legal advice.

A more detailed summary of the draft version of the statement can be found in our post here.

Herbert Smith Freehills Kramer LLP contributed to the consultation. 

Key areas addressed by the UKJT and conclusions reached

The core focus of the statement is whether and in what circumstances those who have not set out deliberately to cause harm may be liable for harms resulting from the use of AI. In this context, the statement addresses the following areas.

Contract and negligence

• Contract will generally be the primary basis on which liability for harm is allocated amongst parties within an AI supply chain, in which case the question of liability will depend on the terms agreed subject to the normal rules limiting freedom of contract. 
• In the absence of a contract, in most cases the question of liability will be determined by reference to the law of negligence. The established legal principles will be applied to the negligence analysis, adapted if necessary to novel factual situations. 

Vicarious liability 

• As AI does not have legal personality, no one can be vicariously liable for the actions or failures of an AI system itself. However, an employer can be held vicariously liable for AI-related harm if that harm arose because a human employee acted wrongfully while using AI.

Professional liability 

• A professional’s obligation to exercise reasonable skill and care in carrying out their duties extends to their use of AI.
• Judged against that standard of reasonable skill and care, a professional may be found negligent for using AI inappropriately, using an unsuitable model, failing to conduct proper due diligence or failing to test AI or validate its outputs effectively. A professional could also be liable for failing to use AI.

Liability in the absence of fault

• The general position is that the risk of harm lies where it falls in the absence of fault. However, there are exceptions, such as the statutory imposition of strict liability for death, personal injury or damage to private property caused by a defective product regardless of whether the manufacturer was at fault. As the law currently stands, this would only apply where AI was incorporated into a physical product.

Causation

• In general, it should be possible to apply the standard "but for" test for causation in the context of AI. To the extent challenges arise, these are not unique to AI and English law has a track record of developing mechanisms to address them.
• The question of whether someone who develops AI can be liable for its misuse by a bad actor or for autonomous acts of the AI will be highly fact-specific. However, someone who develops or deploys AI is unlikely to be held liable for misuse by a bad actor unless the AI in question was obviously dangerous, or the person could have prevented the misuse but failed to do so. However, it is likely that a developer/deployer of an AI system would be liable for harms caused by the AI acting autonomously, unless acts of the kind in question were unforeseeable.
• In principle, the partial defence of contributory negligence is available to reduce damages. A non-commercial user of AI would generally be less likely to be found contributory negligent than a commercial user, but this will depend on the factual context. 

False statements made by AI chatbots 

• A person may be liable for a statement in law if it is made negligently or fraudulently, or if the statement is defamatory. Liability for deceit could also arise, but only if the developer or user intended AI to produce the statement and for someone to believe it was true. 
• Liability would generally be established if a legal person held out an AI chatbot as communicating on their behalf, or if there is an express or implied representation that the chatbot’s statements are correct. 
• Liability can also arise for defamatory statements made by AI, but only for a person deemed in law to be a publisher of the AI’s output. The extent to which those within the supply chain will be considered publishers will be fact-specific, but those who exercise manual review over the output before it is published will be liable as editors.
• The context of statements (eg clear warnings that words are AI generated and may contain hallucinations) will influence the extent to which they are defamatory.

Scope of the statement

The UKJT notes that the purpose of the statement is to clarify the likely application of the law to novel scenarios in circumstances where it may take a long time for cases in these new areas to come before the courts. However, it acknowledges that a comprehensive analysis of the legal terrain is not possible, nor can the statement give definitive answers to the question of whether and how liability arises for AI harm in any particular situation. Instead, the focus is on generic scenarios of wide application. 

A number of areas are specifically out of scope. These include various bases of legal liability (such as economic torts and competition claims) and areas of law such as the interaction between AI and IP rights, data protection, and issues that arise specifically out of the use of AI by public authorities. Legal issues that are not concerned specifically with harms, such as how the law of contract formation applies in the context of AI, are also not addressed.