01What happens.
Administration, dosing and instructions for use exist in several versions across a pharmaceutical company's estate, and AI has no way of ranking them by date. The company's revision process reaches the core site. It does not reach the event microsites, old programme pages, PDFs on partner routes and regional variants where the older versions remain.
02Why it matters.
Medical affairs published the revision correctly, and the instructions on the main site are the instructions. What a clinician now meets first is an AI answer selected from the wider estate, and the version that answers is the one the machine found most readable, most linked or most repeated, whatever its date.
In a published study, over a fifth of AI answers to risky drug questions were judged capable of causing severe harm or death (npj Digital Medicine). A wrong dose administered on the company's own superseded words is the sharpest form of that risk and the company's name is on the answer.
03The case.
This illustrative scenario draws on patterns observed across pharmaceutical estates, see use case 16.
A pharmaceutical company revises the administration guidance for one of its medicines. The summary of product characteristics is updated, the patient leaflet is reissued, the healthcare professional pages on the main site are rewritten, and the change is communicated through the usual channels. Inside the company the revision is complete..
The old guidance survives in 4 places. A microsite built for a launch programme three years earlier still carries the original dosing table, and a distributor in another region hosts the original PDF on its own site. A regional subsidiary's website was translated from the original and never re-translated. A conference page from the year of launch links to the original guidance document, which is still in the document store.
A clinician asks AI how the medicine is administered. The answer comes from the launch microsite, which is well structured, specific and cited by the conference page and the distributor. It states the original schedule with confidence, and it does not say that the schedule has changed, because nothing on that page says so..
The clinician follows the answer, and the dose administered is the old dose. Where the revision was made for safety, the patient receives the dose the company had decided was unsafe, on the strength of the company's own published words.
Medical affairs learn of it through a pharmacovigilance report. The search for the source finds the microsite, then the distributor's PDF, then the regional site, then the conference page, and each is retired or corrected in turn. Each had been outside the process that governed the revision, because the process covered the documents on its inventory and the inventory stopped at the core site.
04Exposure.
The company is of course accountable for clinical instruction published under its name, wherever it sits, and a dose administered on the company's superseded guidance is a matter for the company's product liability, its licence and its directors. Directors are expected to know what clinical instruction the company has in circulation and to take reasonable steps to control it.
Whether any duty is engaged on the facts is for the company's advisers, and the position is stated here as accountability rather than breach.
The consequence begins with the patient. It continues through the adverse event report and the regulator's questions about how superseded instruction reached a clinician. It ends with the correction of the properties the company controls and the request to the distributor to correct the fourth.
05What AI Readiness changes.
The revision process reached the documents on its inventory, and the exposure sat in the ones the inventory did not hold. What was missing was the view from outside: each version of the instruction AI is drawing on, on the company's own properties and on partner and regional routes, with the version AI serves for each question. AAAnow provides that view and evidences it.
On it, the company establishes the current instruction for each medicine in each market where it applies. It retires the versions it controls that contradict that instruction, has partners correct the versions they hold, and checks the answer a clinician receives before the next revision..
Your team makes the changes; AAAnow provides the assessment, the evidence, the guidance and the review.
06Related use cases.
A withdrawn medicine still carried full dosing guidance online, and AI served it as current
The documented origin of this exposure.
Patients were told a withdrawn medicine was available, because the website still said so
Withdrawn products kept evident and citable.
07References.
- npj Digital Medicine, 2024: over a fifth of AI answers to risky drug questions judged capable of severe harm or death. www.nature.com/articles/s41746-024-01346-6nature.comWe do not enable external links - please copy and paste, you are then certain as to the link being opened
- Illustrative scenario drawn from patterns observed across AAAnow assessments of pharmaceutical estates.
Next step
Talk it through, in your terms.
The briefing sets this case in your organisation’s terms, and the discussion follows your questions.
Illustrative scenario, drawn from observed patterns. No company is identified.