The UK Jurisdiction Taskforce (UKJT) has published its Legal Statement on Liability for AI Harms under the private law of England and Wales, concluding that English law is largely capable of dealing with AI-related harm without the need for a bespoke AI liability regime.
The statement aims to provide greater certainty for businesses, developers, professional advisers and users grappling with the question of who is liable when AI causes harm.
A central theme of the UKJT’s analysis is that the absence of AI-specific liability legislation in England and Wales does not leave victims without remedies. Rather, established legal concepts such as contract, negligence, product liability, misrepresentation and defamation are generally capable of being applied to AI-related disputes. As a result, it does not believe that AI requires a fundamentally new liability framework. Instead, the statement argues that English common law’s flexibility allows it to adapt to novel technologies in much the same way it has adapted to previous technological developments.
The statement adopts a deliberately technology-neutral definition of AI, defining it as “a technology that is autonomous”. It identifies three key characteristics that create legal uncertainty:
- an unpredictable relationship between inputs and outputs;
- opacity in the system’s reasoning process; and
- limited user control over outputs.
The UKJT considers these autonomous features, rather than any particular technical architecture such as machine learning or large language models, to be the legally significant attributes of AI.
The statement emphasises that contract will remain the primary mechanism for allocating AI-related risk within supply chains. Relationships between foundation model developers, application developers, hosting providers and enterprise users will usually be governed by contractual terms dealing with warranties, indemnities, limitations of liability and risk allocation. For many AI vendors and customers, the most important liability questions will therefore continue to be answered in commercial negotiations rather than the courts.
Where contractual remedies are unavailable, negligence is likely to be the principal route to liability. The UKJT concludes that there is “no conceptual reason” why negligence principles cannot be applied to AI harms. However, outcomes will be highly fact-sensitive. The statement suggests that:
- careless users of AI are likely to be liable for foreseeable harm;
- developers of narrow, task-specific AI applications may often owe duties of care; and
- foundation model developers will generally be less exposed where harm results from unforeseeable or inadequately tested downstream uses of general-purpose models.
The statement contains particularly important guidance for lawyers, accountants, surveyors, architects and healthcare professionals.The UKJT concludes that professionals may be liable not only for negligent use of AI, but also for failing to use AI where a competent professional would have done so. Examples cited include:
- using unsuitable AI tools;
- failing to carry out appropriate due diligence;
- inadequate testing and validation of AI outputs;
- failing to supervise or verify outputs; and
- failing to adopt AI tools that have become standard professional practice.
The statement also highlights recent examples of lawyers submitting AI-generated hallucinated authorities to courts, making clear that reliance on AI does not displace professional responsibility.
The UKJT firmly rejects any suggestion that AI systems themselves can be liable. AI has no legal personality under English law and cannot owe duties or incur liability in its own right. Consequently, no one can be vicariously liable for an AI system’s actions as such. However, employers may still be vicariously liable for AI-related wrongdoing by employees who use AI negligently.
The statement also considers a recurring question in AI policy debates: can liability arise without fault? The answer is generally no, except in limited product liability scenarios under the Consumer Protection Act 1987. Crucially, the UKJT concludes that the current strict liability regime typically applies only where AI is embedded within a tangible product. Pure software and most standalone AI services remain outside its scope, but that position could change. The statement notes the Law Commission’s ongoing review of product liability rules.
The statement also addresses misleading chatbot outputs. The UKJT concludes that liability for negligent misstatement, deceit or defamation may arise where a legal person can be said to have adopted, authorised or held out AI-generated statements as its own.
