AI developer liability for harms linked to autonomy of their systems

The link between AI autonomy and liability for harms was referenced by the UK Jurisdiction Taskforce (UKJT) in a final legal statement it has issued on liability for AI harms under the private law of England and Wales. The UKJT is chaired by leading judge Sir Geoffrey Vos and made up of a wider panel of barristers, lawyers and academics.

The UKJT’s statement is non-binding, but experts in technology contracts and disputes at Pinsent Masons said it offers greater legal thinking and predictability for both AI developers and businesses that deploy AI on the circumstances in which risks of civil liability under English law might arise from use of AI. This, they said, is because there have been few cases to-date in which liability for AI-related harms has been examined by the courts. 

David McIlwaine of Pinsent Masons said: “A core theme of the legal statement is the rejection of the idea that AI has created a legal vacuum and operates outside the existing law. English law has a well-established track record of developing mechanisms for addressing technological disruption and existing doctrines in contract, negligence, professional liability, product liability and causation can generally accommodate AI-related disputes.”

“The statement highlights two broad sources of liability for loss caused by the use of AI, being voluntary responsibilities primarily imposed by contract and liability imposed by law, commonly assessed under the principles of negligence. The statement consistently reminds readers that AI is not a legal person and therefore focuses on human responsibility and decisions around the AI, rather than the systems themselves, including the importance of testing systems and understanding limitations and bias,” he said.

Laura Gallagher, also of Pinsent Masons, added: “Whilst the statement is not intended to be legally binding, it is clear that this industry-led initiative is likely to be highly influential in how the courts and lawyers grapple with AI harms in the future.”

According to the UKJT, there are a range of circumstances in which businesses or individuals could be liable for AI harms under English private law, even if they do not set out deliberately to cause harm.

For example, it explored scenarios in which employers could be liable for harms caused by employee AI use; in which professionals, like doctors, lawyers, auditors and architects, might be liable for AI harms owing to a failure to exercise reasonable skill and care in carrying out their duties, highlighting that the standards around what constitutes ‘reasonable skill and care’ in an AI context will evolve over time; and in which strict liability for AI harms might apply regardless of fault, such as under product liability rules. Liabilities for AI harms could arise under defamation law too, the UKJT said.

One focus of the legal statement was on when AI developers might face liability for harms arising from others’ use of their models or systems. The UKJT considered the autonomy of an AI system to be relevant to a developer’s liability for related harms.

“It is highly likely that a developer and/or 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,” the UKJT said. “Much will turn on the capabilities and limitations of the AI in question, the level of autonomy and the degree of supervision that is exercised or should have been exercised over it.”

The UKJT said whether liability arises will depend in large part on the facts in each case. It highlighted how developers may have a partial defence against claims of liability if it can evidence “contributory negligence” by third parties.

“Where someone who suffers harm has themselves been at fault and where there is a causal link between that fault and the harm that they have suffered, a court may reduce the damages that the person would otherwise have been awarded ‘to such an extent as the court thinks just and equitable having regard to the claimant’s share in the responsibility for the damage’,” the UKJT said. “The extent of the reduction depends on both the degree of fault on the part of the harmed person and the degree to which that fault contributed to the harm suffered. Contributory negligence is never a complete defence. It can reduce but not extinguish liability for harm.”

“There is no reason in principle why the partial defence of contributory negligence cannot apply to AI harms, particularly where the person who suffers harm is using AI in a commercial context. Examples of scenarios where this might arise include a user failing to follow the instructions and warnings supplied by an AI system or failing to intervene when errors first emerged,” it said.

In its statement, the UKJT also flagged a link between the generality of AI models that developers build and the likelihood of their liability for loss arising from others’ misuse of those models.

“In general terms, the conclusion of the legal statement is that someone who develops or deploys AI is unlikely to be held liable for the misuse of the AI by a bad actor unless the AI in question was obviously dangerous, or the person had the power to prevent the misuse but failed to do so when it should have done,” the UKJT said. “The more general purpose the AI in question, the more extreme the circumstances would have to be before liability would arise.”