TL;DR

The government-commissioned review of English maternity care led by Baroness Amos used Microsoft’s Copilot to help process more than 10,500 submissions, the Daily Mail reported on 1 November, citing an interview Dr Bill Kirkup gave to the Sunday Times. Kirkup, who quit the review team, says advisers saw only AI output, and that references to ‘normal birth ideology’ never reached them. Whatever the outcome, it is a pointed test of how public inquiries use AI on sensitive testimony.

The dispute turns on a sentence in the review’s June report: “We did not find that normal birth ideology was currently widespread.” Mothers from Oxford University Hospitals Trust say their evidence pointed the other way. According to the Mail, 40 pages of accounts from that trust mentioned the phrase twice on the first page alone, and women described being told to aim for a normal birth “at all costs”.

The competing accounts

Kirkup, an obstetrician, is the principal critic. He told the Sunday Times that “advisers never saw those transcripts or the written evidence submitted by families”, adding: “The work was entirely based on the output from the AI tool.” He also alleges that when the issue came up anyway in expert meetings, it was ignored. His own resignation was over this same failure to recognise the ideology, so he is not a neutral witness.

Rebecca Matthews, a founder of the Oxford mothers’ group, says the harm there “has never been properly analysed”. Former health secretary Jeremy Hunt called the possibility “deeply troubling”.

The Mail says Baroness Amos gave no response when approached. Speaking to MPs in July, she said the team did not “find evidence of normal birth ideology”, and pointed instead to failures to recognise and escalate problems.

We have not seen the Sunday Times original, the transcripts or the summaries, and this article relies on the Mail’s account. Nor does the report establish how the tool was configured, what prompts were used, or whether a person checked its summaries against the source material. Those are the questions that would settle whether the software is to blame.

Why it matters beyond maternity care

An inquiry handling that volume of evidence has an obvious reason to reach for a summarisation tool, and other public bodies facing large volumes of responses will recognise the appeal. In our view, the risk is that a summary can read well and still lose a recurring theme, particularly one expressed in the varied language of personal testimony.

Looking Forward

In our view, the practical lesson for any inquiry, regulator or consultation team is about method rather than a ban. If an AI tool sits between evidence and decision-makers, people who gave that evidence will reasonably ask how it was checked. A published account of the method, and a sample audit against the original submissions, would be a modest safeguard. Whether ministers or the review will respond to these claims remains to be seen.