Should patients help build health AI? Nine proposals from the Cercle P and the questions they leave open

In November 2024, a French think tank asked 58 patients and patient representatives about artificial intelligence in their care. Seventy per cent of the patients said they had never used AI in their own care pathway, and that the technology was not among their main concerns. Yet 88% of them, and 82% of the representatives, said patients should be systematically informed whenever AI is used in their treatment. A separate survey relayed by MaPatho, a patient-founded platform, drew a tenth of the usual responses, and described AI as anything from a connected soft toy to a diagnostic aid.

The think tank is the Cercle P, which Pfizer France founded in 2016 to reflect on the place of the patient in the health system. Its 2025 report, "L'IA en santé : les patients ont leur place !", sets out nine proposals in three chapters: inform patients, involve them in building and monitoring AI tools, and mobilise them, their representatives and public authorities to support adoption. Between January and February 2026, three members each presented one chapter on Tech Anatomy: Jean-François Pomerol, CEO of the digital pathology company Tribun Health; Lydia Morlet-Haïdara, a lawyer who directs the Institut Droit et Santé at Université Paris Cité; and Dr Grégoire Pigné, an oncologist and radiation therapist who founded PulseLife, formerly 360 medics. A disclosure is due: Anca Petre, who hosts Tech Anatomy and signs this article, was a member of the group that wrote the report, and MedShake Studio also produces a podcast for Pfizer.

Information beyond the legal minimum

Pomerol's starting point is that most information about health AI is written to comply with the law. It is accurate, he says, but rarely read and of little use to a patient who wants to know why a tool is being used in their own case. Patients ask about AI, in his view, because it is new and carries fears about data, bias and a medicine that might become "dehumanised".

The report answers with three levels of information. The regulatory level stays. Above it, learned societies, the specialty organisations that issue clinical guidelines, would explain what AI is useful for in their field and set guidance on its use, with the Agence du Numérique en Santé, the French digital health agency, helping through labelling. Closest to the patient, the clinician would explain in a few sentences what the tool did in this case. Because consultations are short, the report pairs the conversation with written guides, like the booklets that already help people live with cancer. The report also proposes that hospitals and doctors present their use of AI as a reason to choose them.

What the law already requires

Morlet-Haïdara's chapter starts from a paradox. The legal framework protects patients' privacy and data well, she says, but it has largely forgotten "démocratie sanitaire", the principle introduced by the 2002 Kouchner law on patients' rights that patients should take part in running the health system, and not only be protected by it.

On information, she reads the texts as thin. Since the 2021 bioethics law, Article L4001-3 of the French public health code requires a professional who uses a medical device with an algorithm trained on large datasets to ensure the patient has been informed, without saying who informs, when or how. The EU AI Act's Article 13 obliges providers of high-risk systems to inform deployers, which in health means hospitals and clinicians rather than patients. Article 50, a general transparency rule numbered 52 in the Commission's original proposal, requires that people be told when they interact directly with an AI system, which she considers too imprecise for care. She also cited an industry survey suggesting that many professionals who use AI do not mention it to their patients; we could not trace that survey to a published source.

Her proposals follow from this. Volunteer patients would act as beta testers of AI tools before and after they reach the market, recruited through Mon espace santé, the national personal health account, and listed on a platform that developers could consult. A form of early access, borrowed from medicines, would let a tool with a sufficient level of evidence be tested on a patient cohort, with the results used to set its price and decide on reimbursement. Consent to the reuse of data would be collected once, in one place, rather than for each new study. She pointed to a judgment of the EU Court of Justice of 4 September 2025, in EDPS v SRB, which held that pseudonymised data may not be personal data for a recipient who has no reasonable means of identifying the people concerned, while it remains personal data for the party holding the key. In her reading, this gives researchers more room.

Adoption and the role of the state

Pigné places the subject inside what he calls an unsolvable public health equation: more elderly patients with several conditions, a shortage of doctors, spending at its ceiling, and innovations that add cost. In that context, he says, AI is not yet a political priority.

Patient representatives, he notes, are volunteers whose time is already stretched. The report proposes training for those who want to become "ambassadors", a public consultation on AI in health, and greater representation of patients with AI expertise at the Haute Autorité de Santé (HAS), the French health technology assessment body. He also raised a risk on the professional side: clinicians who use large language models with confidence but without knowing their limits, including hallucinations, which he described as "credible errors" that are hard to spot. His last argument was industrial. France is a small market that cannot pay high prices, so French AI companies need public procurement, and fast, to reach the volume that lets them grow in Europe.

Two readings of the same proposals

One reading treats patient participation as a condition of usefulness. Tools designed without their future users risk being underused or abandoned, as Morlet-Haïdara argued about connected devices for chronic disease, and patients can flag malfunctions after launch, which matters more for systems that keep changing.

A second reading looks at who is speaking and in which direction the proposals point. The report was convened by a pharmaceutical company, and several of its members lead companies that sell digital or AI tools. Most proposals aim to accelerate adoption: presenting AI as a reason to choose a hospital turns information into a form of promotion, early access trades some evidence for speed, and the procurement argument is industrial policy as much as patient policy. A reader who shares the goal of patient participation can still ask whether patients recruited as testers and ambassadors keep the freedom to conclude that a tool should not be used.

What the report does not establish

The figures that open the report come from a survey of 58 people at a single Pfizer patient event and from five qualitative interviews; they describe a well-connected group, not French patients at large. The MaPatho survey is cited mainly for its low response rate. None of the nine proposals has been tested, and the report does not estimate their cost or say who would fund the volunteer platform, the training or the early access cohorts. The reach of the 2025 Court of Justice judgment for health data is still debated among lawyers.

What to watch

The proposals lend themselves to simple checks: whether a French learned society publishes guidance on AI written for patients rather than for clinicians, whether Mon espace santé ever asks its users to volunteer for testing AI tools, and whether the HAS adds patient members with AI expertise to its committees. The Cercle P's earlier proposals on patient teachers and personalised prevention reappeared in French regulation years later, one of them with weaker consent safeguards than the group had asked for. Whether patients end up as co-designers of health AI, or mainly as its early adopters, may depend on which of these proposals are taken up and in what form.

Listen to the episode

IA en santé : les patients ont leur place ! avec Jean-François Pomerol, Cercle P

Listen · 25 min