BMJ Future Health
October 2026
BMJ Future Health Europe is this month, where our community of more than 700 attendees will come together for two days. Key themes this year are AI, women in leadership, digital transformation, and patient-shaped innovation.

We’d love to see you there if you have a ticket, and if you don’t yet, there’s a discount code at the end of this newsletter.

In this issue: two researchers share their perspective on accountability when clinical AI gets it wrong; our BMJ Digital Health & AI Connections journal publishes its first issue; and a paper in The BMJ asks whether digital menstrual tracking risks more harm than benefit.

 

Who’s accountable when clinical AI gets it wrong?
We asked James Siddle and Alistair Connell to share their perspective on the practical implications of their editorial “Shifting bottleneck: AI capability and healthcare adaptation,” published in BMJ Digital Health & AI.

When AI contributes to a wrong clinical decision, who should be responsible, now and in five years?

  • Today, accountability defaults to the clinician who acted on the output. That is defensible only while AI is genuinely advisory and its outputs are checked, and it is already under strain. Ambient scribes [tools that use generative AI to record and summarise consultations into structured clinical notes] ask clinicians to vouch for notes they did not write. Worse, patients who turn to general-purpose chatbots carry the risk themselves.
  • Within five years, responsibility should follow control. Manufacturers should answer for performance within their products’ intended use, deploying organisations for how tools are configured and monitored, and clinicians for the decisions genuinely left to them. Clinicians may carry less responsibility for individual patient-facing decisions, but more for systemic oversight. Where those lines fall is a value judgement. That is why we argue for a citizens’ assembly to settle these questions, rather than leaving the courts to work it out after harm is done.
For someone wanting to innovate with 'low-risk' AI tomorrow, what's the first thing you'd automate?
  • Start where mistakes are visible. Clinical coding is an obvious first pick: it is high-volume, labour-intensive, and every output can be audited against the record. Regulatory compliance and governance are another. They are notoriously slow and burdensome. In both, errors can be surfaced and corrected, and organisations learn to govern AI before the stakes rise.
What would you avoid?
  • What we’d avoid for now is autonomous AI making clinical decisions with no safeguard in between, whether advising patients directly or acting within workflows. Patient-facing failures are especially insidious: a wrongly reassured patient simply doesn’t seek help, and we may never learn how or why the system failed. This isn't because autonomous clinical AI lacks potential; we believe it has a great deal. It is because the trust layer it needs does not yet exist. 

The trust layer, as proposed by the researchers, is a system that would check every clinically consequential output against clinician-approved escalation rules before it reaches a patient.

James Siddle is principal consultant at Emergent Systems and medtech consultant at Our Future Health. Alistair Connell is director of digital health at Our Future Health and president of the digital health section of the Royal Society of Medicine. They have no competing interests to declare.

Meet the editor: Felix Holl's latest articles. 

BMJ Connections Digital Health & AI, an open access journal, publishes original research, including negative results and small or specialist studies, spanning medicine, computer science, public health, and related fields.

Ambient voice technology was among your most-read topics the past month, so this may be of interest: “Effects of an ambient AI scribe used to document patient encounters in hospital: a prospective, multisite, multidisciplinary before-and-after cohort study”, a multisite study putting one of the fastest-growing clinical tools to the test.

The journal's editor, Felix Holl, deputy head of the DigiHealth Institute at Neu-Ulm University of Applied Sciences, told us back in February why he wanted to lead the journal: “I’ve seen too many great digital health and AI ideas die in pilots or fail in real-world care... We need fewer demos, more outcomes.” Read the full Q&A →

Got a question of your own for Felix? Send it our way. If there’s interest, we’ll put a fresh set to him for a future issue.

From across the BMJ Group

We selected these papers from BMJ Group journals to help you keep pace with the issues that arise at the intersection of technology and healthcare. 

Review Editorial Analysis Opinion

From the community editor

Patients can get health advice from influencers, podcasters, and AI chatbots before they ever see a clinician, and many of those sources are commercial.  We’re exploring this on a panel at BMJ Future Health Europe on 29 October. I’ll be joined by clinician content creators Aziza Sesay and Ally Jaffee, chaired by The BMJ’s Rebecca Coombes. Join us if you’re attending.

All newsletter subscribers receive a 20% discount to attend BMJ Future Health Europe. Use this code at checkout: FHNEWS20.

Thank you for reading. I look forward to seeing some of you at BMJ Future Health Europe.

Loading