The EU AI Act’s countdown clock hits zero today — and most of this edition is, one way or another, about why the machines still need minders.
EU AI Act goes live today — and EMA Annex 22 lands the same day
Two regulatory deadlines collide today: the EU AI Act takes full effect, and EMA Annex 22 — the first framework explicitly governing AI in drug production — kicks in alongside it. If your organisation touches EU submissions or regulatory communications, the compliance clock has stopped counting down and started counting. Documentation and disclosure language that isn’t ready now is officially late.
Fake drug names fool LLMs — affixes alone trigger confident clinical responses
Give a large language model a fictitious drug with a plausible suffix — “wugcillin”, say — and it will cheerfully generate pharmacology for an antibiotic that does not exist. For health communications teams using AI to draft drug-related content, this is a documented hallucination vector that sails past the obvious red flags, because nothing about the output looks wrong. The lesson is not new, but the mechanism is: verification has to happen at the level of the compound, not the prose.
AI evaluators drift with conversation mood — your review pipeline is compromised
If your team batch-reviews promotional copy or MLR-prep outputs through an LLM in a single long conversation, earlier items are silently skewing the scores of later ones — negative content by a factor of 1.6. The fix is almost embarrassingly simple — fresh context per item — but most off-the-shelf review workflows don’t do it. Worth an hour of someone’s Monday to check whether yours does.
Google’s AI search overhaul buries web results under ads and summaries
Google’s expanding AI Overviews and conversational ads are designed to keep users on-platform, which means pharma-funded content, medical publishers, and HCP-facing web properties see less organic traffic. If your evidence dissemination strategy still assumes people click through to websites, that assumption is eroding quarter by quarter. Distribution is becoming a question of what the machine chooses to summarise — and whether your content is in the summary.
The intern and the blank page: how medical writers are actually living with AI
A worried student asked working medical writers whether AI is coming for the profession, and the answers are more useful than most conference panels. One treats AI as a brilliant intern whose every output needs checking; another keeps it out of first drafts entirely, on the grounds that the blank page is part of the craft. The split itself is the story — and both camps quietly agree on the uncomfortable part: teams will shrink as one writer absorbs more of the output.
That’s it for this edition. Back Monday.
— Ned
Follow the conversation: #IrreplaceablesHealth