Skip to main content
Share via Share via Share via Copy link

AI, data and the law in health and life sciences: Who owns it, who's liable and who decides?

19 November 2026
Gerard Hanratty and Clive Kay

Artificial intelligence is moving into healthcare faster than the legal and regulatory frameworks designed to govern it. 

Across diagnostics, clinical decision-making, research and operational management, AI tools are being deployed in settings where the questions of ownership, accountability and oversight have not yet been resolved. 

Who owns the data generated when a private technology company processes information from a public health system? Who is responsible when an AI system contributes to a clinical decision that goes wrong? And who has the authority to determine that a tool is safe and appropriate to use in the first place? These are not abstract questions for the future, but live issues that health organisations, clinicians, lawyers and regulators are navigating right now.

It was against that backdrop that Browne Jacobson hosted a satellite event as part of London Life Sciences Week at Somerset House, bringing together clinical, digital, legal and innovation perspectives to work through these challenges. 

The session was chaired by Professor Clive Kay, Strategic Adviser to Browne Jacobson’s health and life sciences team, with panellists including:

  • Gerard Hanratty, Head of Health and Life Sciences, Browne Jacobson.
  • Dr Rishi Das-Gupta, Chief Executive Officer, Health Innovation Network South London.
  • Beverley Bryant, Chief Digital and Information Officer, University Hospitals Dorset NHS Foundation Trust.
  • Professor Bu'Hussain Hayee, Consultant Gastroenterologist and Clinical Director for Liver, Gastroenterology, Upper GI and Endoscopy, King's College Hospital NHS Foundation Trust.

The purpose of the discussion was not simply to celebrate the technology or to treat regulation as an obstacle. Questions were instead presented on how organisations can make sound decisions, support innovation and protect patients – all at the same time.

A clinical lens: AI in the endoscopy suite

A vivid illustration of AI's real-world impact was presented to delegates. Bowel cancer is the second most common cause of cancer death in the UK, yet adenoma detection rates remain stubbornly below 40%, with significant variation across the country. 

AI-assisted endoscopy – specifically computer-aided detection, or CADe – has been trialled in clinical settings to flag polyps at speed and consistency that human operators alone can’t match across a full working day. 

Between November 2023 and February 2025, CADe analysed 4,514 colonoscopies from 25 hospitals across England, Wales and Scotland. Adenoma detection rates increased from 30% to 34.8% when AI was active. 

Crucially, the study found no evidence of clinicians being de-skilled by working alongside these tools. 

It was noted, however, that as AI moves along the spectrum from detection to diagnosis to quality assurance to full clinical decision support, the ethical and legal stakes rise with every step. This case study provided the backdrop for a wider discussion on the use of AI and big data in healthcare.

Who owns the data?

Ownership of health data generated in NHS settings is, in one sense, settled – patients own their data. But as one panellist observed, most patients have little idea what data is held about them, let alone how it is being processed or shared. The legal framework of controller and processor still applies when AI is used; the tool changes, but the accountability does not.

What is less settled is the question of value. The UK, with its state-based healthcare service, holds one of the most comprehensive structured health datasets in the world – a genuine national asset for tackling cancer, diabetes, and a range of conditions that affect populations globally. The challenge is that this value is not being realised. One example cited in discussion was a bed-tracking system developed with significant public investment being virtually given away to Australia rather than fully commercialised – something, it was noted, that would not have happened in the United States.

The consent question is also unresolved. There’s a belief that patients will generally support AI being used to improve their own care. The concern often arises when they understand their data will form part of large-scale datasets and where their data may end up. But the benefits of large-scale anonymised datasets were evident in the COVID-19 vaccines, where they were synthesised using AI for rapid testing and rollout in less than a year, as opposed to the typical 10 to 15 years it can take to develop and receive authorisation.  

Separating those two conversations – individual care and population-level research – is essential, but frameworks don’t currently enable this in a coherent way. The analogy of organ donation's shift to an opt-out model was raised as a possible template for thinking about data governance differently.

Who is liable when things go wrong?

The honest answer, one participant observed, is 'potentially everyone'. Manufacturers, deploying organisations and clinicians may all carry some exposure depending on how an AI system is configured, used and monitored. 

The hypothetical example of an agentic AI system that, without anyone intending it, began prioritising patients based on the first letter of their surname illustrates the problem vividly: the manufacturer says it was not programmed that way but the deploying organisation says it was not warned. The courts will ultimately be required to draw the lines. 

A more subtle liability risk was also raised regarding the failure to use AI at all. With long waiting lists and cancers being diagnosed late, there may come a point at which not deploying available tools is itself a cause of harm. 

Who decides, and how do we govern this well?

There was broad agreement that the current procurement and governance landscape is not fit for purpose. Every NHS trust independently assessing AI tools represents a failure to use the collective buying power of the system. 

Faster, more centralised frameworks, including innovation passports that allow one trust's evidence of efficacy to be recognised by others, could make a material difference.

Regulation, too, needs to move at the speed of the technology. The EU AI Act's static approach was flagged as a cautionary tale. The MHRA's emerging lifecycle approach, including AI sandboxes, is more promising – but the consensus was that the regulator needs to move faster. 

Post-market surveillance, in particular, needs to be more rigorous. Approval at one site does not guarantee safe performance at another, given the variability between NHS systems and patient populations.

One audience member, reflecting on the experience of the COVID-19 vaccine rollout, made a pointed observation that innovation moved so quickly in part because many governments agreed to indemnify or underwrite liability risks, allowing companies like AstraZeneca to rapidly manufacture and distribute it on a not-for-profit pandemic basis. 

This is not a sustainable or scalable model, however. What is needed instead is a more sophisticated, granular understanding of what acceptable failure looks like – distinguishing between risks that are truly intolerable, such as data breaches, and those that are manageable with the right monitoring in place.

Aligning law, governance and public acceptance 

The technology is ahead of the law. The law is ahead of the governance. And the governance is, in places, ahead of the culture. Getting all four aligned – quickly enough to matter for patients, carefully enough to maintain trust – is the defining challenge for health and life sciences in the years ahead. 

You may be interested in...