The AI Liability Gap Is Real. Structure Over Generation Already Closes Part of It
17 June 2026 · 14 min read | Reviewed September 2026 — updated following the National Commission into the Regulation of AI in Healthcare's final recommendations published 10 September 2026
Copy link
Who is liable when AI contributes to clinical harm? There is no single answer. Healthcare professionals remain accountable for their own professional decisions, while healthcare providers, developers and manufacturers may carry different legal and regulatory responsibilities depending on the technology, defect, deployment and circumstances. In September 2026, the National Commission into the Regulation of AI in Healthcare said negligence claims are disproportionately likely to be brought against healthcare professionals and providers because they have the clearest duty of care, contributing to perceived 'liability sinks'. It recommended clearer allocation of responsibilities across the AI lifecycle. ReporticaAI's Structure Over Generation principle does not remove legal liability; it addresses a narrower governance problem by preserving clearer practitioner authorship, provenance and review where AI assists with professional documentation.
In June 2026, Medical Protection published Closing the AI Liability Gap , warning that existing liability arrangements may leave clinicians and healthcare providers exposed when AI contributes to patient harm.
Its intervention identified a genuine policy problem, but the legal position should not be reduced to the proposition that clinicians are always liable and AI developers are not.
Liability depends on the circumstances, the applicable cause of action, the status and function of the AI system, the conduct of the professional and provider, product responsibilities and the causal relationship between the alleged failure and the harm.
Since this article was first published, the debate has advanced significantly.
On 10 September 2026, the National Commission into the Regulation of AI in Healthcare published its final recommendations. The Commission concluded that AI-enabled healthcare requires a system-wide approach to safety and accountability, with distinct but complementary responsibilities for manufacturers, healthcare providers, healthcare professionals, regulators and other system partners.
Crucially, the Commission identified what it called perceived 'liability sinks': situations in which responsibility for errors may be transferred disproportionately onto healthcare professionals and providers without sufficiently recognising the influence of AI systems and wider system design.
That is a more precise description of the liability gap.
The problem is not that developers can never incur liability. It is that existing routes to responsibility and redress may not allocate risk clearly or proportionately across everyone involved in developing, deploying and using an AI system.
Where Does Responsibility Sit When Clinical AI Goes Wrong?
AI does not become the responsible legal actor merely because its output influenced a decision.
Responsibility has to be considered across the people and organisations involved in the AI lifecycle.
The National Commission's September 2026 recommendations expressly reject an approach in which accountability is treated as belonging to only one participant. It says manufacturers, healthcare providers, professionals, regulators and other system partners should have distinct but complementary responsibilities.
This table is a governance map, not a statement that every actor will be legally liable whenever harm occurs. Legal liability depends on the facts and applicable law.
| Actor | Relevant responsibility may include |
|---|---|
| Healthcare professional | Professional judgement, appropriate use, acting within competence, responding to known limitations and decisions actually made in practice |
| Healthcare provider | Safe deployment, governance, training, local assurance, monitoring, escalation and organisational responsibility for care |
| Developer/manufacturer | Product design, intended purpose, performance, safety information, risk management and applicable product or medical-device obligations |
| Procuring/deploying organisation | Due diligence, implementation, contracts, clinical safety and suitability for local use |
| Regulator/system bodies | Applicable standards, enforcement and assurance frameworks |
What Did the National Commission Say About the Liability Gap?
The Commission said negligence claims are disproportionately likely to be brought against healthcare professionals and providers because they have the clearest duty of care. It described perceived 'liability sinks', called for clearer guidance about where professional responsibilities start and end, and said risks should in principle sit with those best placed to manage them.
The Commission expressly says professionals and other AI users should not be asked to carry accountability for risks they cannot plausibly mitigate.
Is the Law Already Settled?
No. The Commission describes liability for AI in healthcare as complex and evolving. It notes that existing common law in England and Wales can address many novel situations while recognising genuine uncertainties. Its September 2026 report is advisory and a cross-government response is to follow separately.
Providers and professionals should distinguish existing law and professional obligations, organisational and product-regulatory responsibilities, current guidance and recommendations for future reform.
What Do Professional Regulators Currently Say?
The GMC guidance on artificial intelligence and innovative technologies states that doctors, physician associates and anaesthesia associates remain responsible for decisions they take when using AI and should apply professional judgement and existing standards. It also recognises that those involved in creating, testing and updating technologies remain responsible for those aspects.
On 17 September 2026, the NMC announced that it is joining the Professional Standards Authority, other regulators and the Accredited Registers Collaborative to develop shared principles for regulating professional use of AI. Until those principles are developed, do not attribute a detailed AI-specific liability rule to the NMC that it has not published.
Structure Over Generation: What It Can — and Cannot — Solve
Structure Over Generation is ReporticaAI's governance principle for AI-assisted professional documentation.
Generation-led documentation asks an AI system to produce substantive professional content or conclusions that the professional then reviews, edits or accepts.
Structure-led documentation begins with practitioner-supplied facts, observations, judgements or source material and uses technology to organise that material into a defined professional structure while preserving a clearer relationship between the practitioner's input and the resulting record.
This distinction does not determine legal liability. A structure-led system can still be used badly; a clinician can enter inaccurate information or fail to review a record, a provider can deploy an unsuitable system and a developer can produce defective technology.
The narrower proposition is about provenance and accountability . Where AI-generated content enters a professional record, later reconstruction may require answering what came from the professional, what the AI introduced, what was reviewed, what changed, who approved the final version and whether the pathway can be reconstructed.
A structure-led architecture can reduce some ambiguity where it preserves source material, separates supplied information from system processing, requires meaningful review and maintains an auditable record.
Why Provenance Matters
Provenance means establishing where information came from and what happened to it. It can support professional accountability, clinical-record integrity, incident investigation, organisational assurance, auditability, identification of AI-introduced errors and reconstruction after an adverse event.
NHS England's current guidance for AI-enabled ambient scribing reinforces the importance of transparency and traceability for accountability and liability where required. That does not endorse ReporticaAI's framework, but it demonstrates why provenance matters.
Structure Over Generation Is Risk Reduction, Not Liability Transfer
These controls can make accountability more visible and documentation more defensible. They cannot decide in advance who would be legally liable if harm occurred.
This is the architecture behind Reportica Pulse's Integrity Trace — explore how provenance is built into clinical placement documentation to create defensible, auditable records where practitioner observation and professional judgement remain central.
Explore the governance framework →
| Governance problem | Structure-led response |
|---|---|
| Unclear origin of professional content | Preserve identifiable practitioner input |
| AI introduces unsupported material | Constrain generation and retain source provenance |
| Unclear whether output was reviewed | Build explicit professional review into the workflow |
| Difficult reconstruction after an incident | Maintain traceable input, processing and finalisation records |
| Automation bias | Keep professional judgement visibly central |
What Changed in September 2026?
The National Commission's final report materially advances the AI-accountability debate.
Its recommendations include clearer roles across manufacturers, providers, professionals, regulators and other system partners; allocating responsibility in principle to those best placed to manage the relevant risk; avoiding accountability for risks professionals cannot plausibly mitigate; clearer guidance about reliance on AI advice; lifecycle approaches to safety and governance; stronger training; improved reporting of adverse outcomes and malfunctions; and greater transparency around AI use in patient care.
These are recommendations for development of the future regulatory framework. They should not be presented as though Parliament or government has already converted every recommendation into a binding legal requirement.
AI-Assisted Clinical Documentation: Accountability Questions
| Question | Why it matters |
|---|---|
| What role does the AI perform? | Responsibility cannot be analysed without understanding function and intended use |
| What information did the professional provide? | Establishes the human source material |
| What did the AI add, transform or infer? | Helps identify the technology's contribution |
| Was the output meaningfully reviewed? | Professional oversight should be demonstrable |
| Who approved the final record or decision? | Clarifies operational responsibility |
| Were limitations communicated? | Relevant to safe deployment and reliance |
| Were users trained? | Competence depends partly on understanding the technology |
| Can an error be traced to its source? | Supports investigation and learning |
| Is performance monitored after deployment? | AI risk can change during real-world use |
| Are incidents and malfunctions reportable? | Reporting supports safety, oversight and learning |
Frequently Asked Questions
Q: Who is liable if a clinician follows incorrect AI advice?
There is no universal rule that determines liability simply because AI was involved. Liability depends on the circumstances, including professional conduct, provider responsibilities, the system and intended use, product obligations, causation and the alleged failure.
Q: Is a clinician always liable for an AI error?
No. Professionals remain accountable for their own decisions, but do not automatically assume legal responsibility for every defect, design failure or risk elsewhere in the AI lifecycle.
Q: Can an AI developer or manufacturer be liable?
Potentially. Developers and manufacturers can have responsibilities relating to design, safety, performance, intended purpose and applicable product regulation. Liability in a particular case depends on the facts and law.
Q: What is an AI liability sink?
The National Commission uses the concept for concern that responsibility may fall disproportionately on professionals and providers because they have the clearest duty of care, without sufficiently recognising AI systems and wider system design.
Q: Are the September 2026 recommendations already law?
No. The Commission is an expert advisory body. Its recommendations propose a future framework and a cross-government response is to follow separately.
Q: What does the GMC say about clinicians using AI?
The GMC says doctors, physician associates and anaesthesia associates remain responsible for decisions they take when using AI and should apply professional judgement and existing standards.
Q: What does the NMC say about AI?
On 17 September 2026, the NMC announced work with the PSA and other regulators to develop shared principles for regulating AI use by health and care professionals. Existing professional standards continue to apply.
Q: What is Structure Over Generation?
It is ReporticaAI's governance principle for AI-assisted documentation. It favours preserving practitioner-supplied information and organising it within defined structures rather than allowing AI to become the independent author. It is not law, a regulatory doctrine or a guarantee of legal safety.
This article aligns with PAIDS™ (Professional AI Documentation Standards) — well-sourced, thoroughly researched, and defensible with verifiable data. Explore the governance framework
Need help choosing?