Which AI use cases in pharma and life sciences are high risk under the EU AI Act?

Annex III point 4(b) covers AI systems intended to monitor and evaluate the performance and behavior of persons in a work related relationship. Scoring every line against the employee's own submission history, and instrumenting a repeat flag rate by employee, is that kind of behavioral evaluation, so this design falls under Annex III. The only carve out, Article 6(3), lets a narrow procedural or preparatory task escape high risk with a documented assessment, but the last subparagraph of Article 6(3) removes that carve out whenever the system performs profiling of natural persons. Scoring lines against an individual employee's history is profiling, so the carve out is not available here: keeping a human auditor as the actual decision maker on any personnel action is a required control, not an exit from Annex III.

Drafting regulated documents for expert review is not listed in Annex III and is not a practice prohibited by Article 5, so the tier turns on the sponsor's role under Article 50. A sponsor that deploys a third party drafting tool has no specific AI Act obligations beyond AI literacy: the Article 50(4) disclosure duty covers AI generated text published to inform the public on matters of public interest, which clinical study reports and regulatory submissions are not. For that sponsor the tier is minimal. A sponsor that builds its own generating system, as Merck (a proprietary platform) and Novo Nordisk (NovoScribe) did, is its provider under Article 50(2) and must mark the synthetic text in a machine readable format, unless the exemption for an assistive function for standard editing applies; drafting whole report sections goes beyond that exemption, so for that sponsor the tier is limited. Quality expectations come from medicines regulation and EMA guidance: the EMA reflection paper expects close human supervision and quality review when AI drafts medicinal product information documents, and makes the clinical trial sponsor, marketing authorization applicant or holder, or manufacturer responsible for ensuring that models and data pipelines are fit for purpose and meet GxP standards and EMA guidelines.

Ranking candidate sites as institutions, by data such as facility capacity and historical trial throughput, is an internal research operations decision and stays minimal risk: it does not by itself decide a natural person's access to healthcare, credit, employment or another Annex III listed area. But the same models often also score and tag named principal investigators on performance, speed and representativeness, as in the Novartis evidence on this page, to help decide which investigator's site gets a trial contract. Annex III point 4 covers AI used to evaluate the performance and behavior of natural persons in a work related context, so a deployment that profiles named investigators, not only institutions, may fall inside that point depending on how the sponsor uses the score. Sponsors should classify each deployment on whether it scores named investigators or only sites, and treat investigator scoring as high risk until a legal review of the specific design says otherwise.

Internal intake, extraction and coding for review by safety staff is not listed in Annex III and is usually minimal risk. A public facing reporting assistant must tell people they are talking to an AI under Article 50. The main obligations come from pharmacovigilance law and good pharmacovigilance practices, which require validated, inspectable processes, and from GDPR rules on health data.

Reviewing internal manufacturing quality data on its own is not listed in Annex III, and pharmaceutical products are not covered by the AI Act's Annex I product safety legislation the way machinery or medical devices are, so a deployment limited to that stays minimal risk. The tier moves to high risk under Annex III point 4(b), monitoring and evaluating the performance of workers, if the design attributes findings to, or is used to evaluate, a named operator: batch records carry operator initials, and the Recordati evidence on this page shows this kind of root cause analysis can surface "differences in operator performance". Keep any per operator finding aggregated or advisory rather than used to evaluate a named person, and treat it as high risk the moment that changes. Good manufacturing practice regulation, not the AI Act, sets the everyday bar for the system itself (see PIC/S Annex 22 below).

Prescreening for research that staff verify is not listed in Annex III and is usually minimal risk. The Article 2(6) exclusion covers only systems developed and put into service for the sole purpose of scientific research and development, so an operational recruitment tool used across a health system usually falls inside the Act. If the software recommends trials to a clinician as a treatment option for an individual patient, it may qualify as medical device software under the Medical Device Regulation; where that needs a notified body assessment, it is high risk under Article 6(1). Processing health records for research falls under GDPR Article 9 and national research rules.

Generating presenter video is not listed in Annex III. Providers of tools that generate synthetic video must mark the output as AI generated, and deployers must disclose deep fakes, such as an avatar or cloned voice of a real executive that would falsely appear authentic (Article 50(2) and (4)). The tier changes only if the same system also decides on access to work or services, which these deployments do not describe.

Article 50(1) requires providers to design AI systems that interact directly with people so that those people are informed they are dealing with AI, unless this is obvious from the context, as it usually is for an internal assistant. An analytics assistant that makes no decisions about people is not a prohibited practice under Article 5 and is not listed in Annex III. It would be high risk only if it were intended for an Annex III purpose, such as assessing the creditworthiness of natural persons (point 5(b)).

An internal drafting and review aid that makes no decisions about people. Article 50 transparency duties apply to generated content: providers must mark synthetic content, and deployers must disclose deep fake images, audio or video. Personalized targeting of individuals is governed mainly by data protection and consumer law rather than the AI Act.

An internal tool that monitors and classifies regulatory publications for staff makes no decisions about natural persons, so it is not listed in Annex III and is not a prohibited practice under Article 5. Staff know they are using an AI tool and its summaries are not published to the public, so the Article 50 duties to inform users and to disclose published generated text add little for the deploying organization. Article 50(2) still requires the provider of a system that generates text to mark its output, in a machine readable format, as AI generated: usually the vendor, but an organization that builds its own summarizer can itself be that provider, which is what puts this use case at the limited tier rather than minimal. Beyond this and AI literacy (Article 4), no specific obligations apply. General model risk and third party rules still apply.

Target scoring and molecule generation are not a safety component of an Annex I product and are not one of the Annex III high risk areas (Article 6), so they do not become high risk on that route. Insofar as the platform and its training are themselves scientific research and development, activity that stays there falls outside the Regulation entirely under the Article 2(6) research exclusion. The resulting drug candidate is separately regulated as a medicine, not as an AI system, through the normal pharmaceutical approval pathway; conventional preclinical and clinical testing validates the AI's outputs before anything reaches a patient.