AI in UK BioTech

Why this briefing

On 29 June 2026 the MHRA Inspectorate published a blog post most of the sector will not have seen: it is now receiving responses to inspection findings that cite guidance which does not exist, invoke the wrong regulatory framework, or appear, in the regulator's own words, "designed to mislead rather than address underlying problems." One documented case turned four hours of review into more than twenty, on a serious patient safety deficiency. This is not an attack on the organisation. It is the organisation's own drafting process, in a document class nobody had put on a technology risk register.

Meanwhile the headline number for the quarter is real and almost entirely one transaction: £2.05bn in UK biotech venture investment, the strongest quarter in five years, of which £1.6bn is a single company's Series B. Strip it out and the underlying recovery is £498m — genuine, but a different story from the one the headline tells.

Four companies sit either side of that quarter. One platform is live and adopted. One is the headline deal's own account of itself, now with two named pharmaceutical partners paying to test the claim. One pioneer of the category no longer exists as a UK company. And one company's own record holds both the sector's clearest verified drug-discovery success and an inconclusive one, from the same underlying technology.

The human moment

An inspection response is the mechanism by which a compliance failure gets corrected: a finding is raised, the organisation responds, the regulator accepts or challenges, the fix happens. When that response is drafted by AI and nobody has verified it, the correction loop breaks at exactly the point where it matters most. The MHRA's own requirement has not changed — a submission must be factually accurate, verifiable, and signed off by someone with authority — but nothing currently obliges anyone to check that the review actually happened before the document leaves the building. The person who signs it is the one holding the risk nobody named.

What's inside

Nineteen pages, written for an executive who knows biotech well and AI not at all:

• A note on the numbers — what is included, what was excluded, and why, including a statistic read in full and excluded anyway
• A record quarter, and its denominator — four named companies' divergent outcomes over two years
• The pressure point — the MHRA's own account of AI-drafted inspection responses
• What the sector is facing, under the 2026 clinical trials rewrite and the ICO's remit
• Five risks, each drawn from a regulator's own account or the sector's own data
• The threat side — a broken correction loop, and a published way past a voluntary biosecurity standard
• The UK legal position, and why there is no rulebook for AI in medicines specifically
• Beyond the UK — what other major medicines regulators have written down
• Seven things that can be done now, none requiring a UK framework to exist first
• Three levels, defined by where accountability sits
• Benefit realisation, and the business case still missing its second variable
• An eight-workstream way in

The discipline behind it

Every figure carries a source. Nine bodies of evidence were found and deliberately excluded, and the exclusions are listed in the document — among them a widely repeated "173 AI drugs in trials, zero approved" claim traced to a blog with no methodology, and a paywalled trade headline claiming the MHRA had backed a global pharmacovigilance framework, which would have been a significant finding and was not verifiable.

Two corrections are worth naming. The sector's most repeated statistic — an 80 to 90 per cent Phase I success rate for AI-discovered drugs — was first reported as unobtainable. The source paper was later supplied, read in full, and the figure confirmed real. It is still excluded: all five authors are Boston Consulting Group employees, and the paper states BCG's Health Care practice funded the work. Separately, an early account of Genomics' Mystra AI platform described it as adopted by "Novo Nordisk and others" — the company's own launch release, read before publication, names BridgeBio Pharma, Relation Therapeutics and AWS, not Novo Nordisk. The claim was corrected before the briefing was built, not after.

Who it's for

A biotech AI or data lead, a regulatory affairs or quality director accountable for what leaves the building under their signature, a board member being asked to approve an AI drug-discovery partnership, or anyone who wants a plain, sourced account of what is verified in this sector and what is a company's own claim about itself. It is written to work for all three.