UK Jurisdiction Taskforce Statement on Liability for AI Harms: the clinical negligence perspective

The UK Jurisdiction Taskforce has published its Legal Statement on Liability for AI Harms. The Statement considers how harms caused by Artificial Intelligence may be addressed under existing UK legal principles and is intended to provide guidance to professionals who use AI, as well as indicators of how claims arising from negligent use may be approached by the courts.

This article considers how those issues may apply in the context of clinical negligence claims.  

Key takeaways:

• Bolam still applies
• Non-delegable duty of care will apply to NHS Trusts using AI
• The need to ensure clinicians understand the technology and provide their professional judgment is key.
• The need for insurers to consider policy wordings as to how AI is expressly considered.

Standard of Care (Human oversight remains essential)

The use of AI in connection with the provision of healthcare is becoming more prevalent. What has remained unclear is how the use of AI may be dealt with in a clinical negligence claim where a clinician has relied on some form of AI when making clinical decisions about treatment and it is later alleged that the clinical decision making was negligent.

At paragraph 62 of the Legal Statement, it is explained that the court will require expert evidence as to what a reasonable professional carrying out a specific task should have done. The applicability of any existing regulatory guidelines will form part of that assessment. The Statement emphasises that the question of what reasonable care and skill entails will be “profession-specific, task-specific, and situation-specific”.

The Statement also notes that failure to carry out proper due diligence, or failure to ensure that the user has a sufficient understanding of the technology, will be important factors. If a professional does not understand the AI tool sufficiently, that is likely to support an argument that they could not have exercised reasonable care and skill. A professional who fails to exercise appropriate oversight of the AI system’s outputs is also likely to be at risk of a finding of negligence.

As a result, clinicians and healthcare providers permitting the use of AI tools should ensure that proper training is provided to users, before these tools are deployed. 

Failure to use AI

The Legal Statement explains that, while much of the discussion to date has focused on how a person may be negligent in using AI, there is also the possibility of a professional being negligent for not using AI. At paragraph 67 of the Statement, the example of a radiologist is used. If an AI system is effective at identifying cancerous tumours, a radiologist may potentially be criticised for not using that AI system, especially if it did not withstand logical analysis not to use it (Bolitho).

That may appear to sit uneasily with the requirement for human oversight, particularly where the AI system may be more accurate than the clinician in a specific task. However, the Bolam test is likely to remain central. The question will still be whether the clinician’s management was supported by a responsible body of professional opinion capable of withstanding logical analysis. 

AI Chatbots and Clinical Judgment

AI chatbots may increasingly form part of some clinicians’ practice when researching possible diagnoses or treatment options. However, such tools are known to carry risks, including hallucinations and inaccurate outputs. The Legal Statement is clear that AI has no legal personality and legal liability will not attach to the AI system itself. Responsibility will remain with the clinician to apply their own professional judgment when reaching a decision about a patient’s condition.

We recommend that clinicians therefore need to ensure that where such tools are being used, they clearly document why they chose to use the tool, what prompt they gave it, and how they applied that output to the specific clinical picture of the patient in front of them, and why they agreed with the output.

Vicarious Liability and Non-Delegable Duty of Care

The Legal Statement is clear that an NHS Trust deploying AI diagnostic tools continues to hold a non-delegable duty of care to patients (paragraph 36).  As a result, the Trust is unlikely to be able to escape liability by arguing the AI tool was defective.  The competent use of AI will remain the responsibility of the Trust who must  ensure that only appropriate tools are deployed and clinicians are adequately trained.  Vicarious liability for employees using AI, will also apply where the employee was negligent in the use of AI.

The ability for the Trust or healthcare provider to later recover from the AI developer will remain an option provided that there is a contractual route (robust indemnities in procurement contracts will be crucial) or via the Civil Liability (Contribution) Act 1978 route.

Causation

In a clinical negligence claim, it is not sufficient for a claimant to prove breach of duty. They must also prove, on the balance of probabilities, that the negligent act or omission caused harm. In many clinical negligence claims involving AI, the Legal Statement confirms the existing causation tests are likely to remain applicable (paragraph 88).

The Legal Statement also confirms that there is “no reason in principle why the ‘material contribution to damage’ approach to establish causation cannot be applied to harms caused by AI” (paragraph 107).  

Nevertheless, the principle of causation remains that if a clinician is found to have acted negligently by using, misusing or failing to use AI, the Claimant will still need to establish what would probably have happened with reasonable non-negligent care.  That may prove difficult if it is hard to replicate what appropriate use of AI would have generated.

The more difficult scenario is where the clinician was reasonable in using AI, but the AI output was wrong and the patient suffered harm. In that situation, a defendant will argue that the clinician was not negligent in relying on the tool and so there was no breach of duty and the claim fails.  

Claimants who have suffered harm, will then need to consider whether the real target of the claim is not the clinician’s decision making, but entities responsible for developing or supplying the AI tool itself.

Product Liability

The above problem may lead Claimants to consider an alternative route of product liability, where breach against a clinician cannot be proven.

This in itself has its own complexities where the current law under the Consumer Protection Act 1987 (“CPA”) focuses on defective ‘products’ and pure software within goods does not comfortably sit within that definition.  As set out at paragraph 74 of the Statement, whether a product was defective may not readily apply to an AI system that is not incorporated to a tangible product.  The Law Commission is intending to review whether the definition of a product within the CPA needs to be reformed to accommodate digital technology including software.  A public consultation on proposed changes is expected later this year but it may be some time before changes are implemented.  For now, based on the current understood definitions,  it seems that a chatbot used by a clinician may not be considered a product, but a radiology equipment with embedded AI might fall within the product liability definition.

Conclusion

In summary, the Bolam test is likely to remain central, particularly on whether use or non-use of AI was consistent with responsible professional practice. However, defendants may also need to be careful to preserve and investigate evidence relating to training, system validation, governance, audit trails, warnings, limitations of the tool and the degree of human oversight applied.

Hospitals and healthcare providers need to have strict control over which AI tools are being procured, what training is being given to staff to ensure they understand the technology to be able to exercise reasonable care.  Given the significance of non-delegable duties and vicarious liabilities for employees, any provider who procures AI tools needs to consider the contractual indemnity position, should that tool cause harm to a patient.

We recommend that Insurers and indemnity providers should be reviewing policy wordings now to ensure AI-related risks are being expressly considered. As the use of AI becomes increasingly embedded within healthcare delivery, there is a real risk of "silent AI" exposures where policies neither clearly cover nor exclude losses arising from AI-assisted decision-making. The UKJT's analysis suggests that liability is likely to be determined by established principles of negligence, vicarious liability, and non-delegable duty, making robust underwriting, risk management and contractual indemnity arrangements more important than ever.

Clyde & Co's healthcare group is recognised for its extensive industry knowledge, offering a range of legal services covering public and private sectors as well as inquests, advocacy, professional regulation, product liability and pharmaceuticals/life sciences. Should we be able to assist you, please do contact one of our experts.