Skip to main content

The medicines business | A question to investigate

Governed AI begins with a decision about its role

The first AI decision is not which model to buy. It is what role an output will play, who is responsible for checking it and what happens when it is wrong. A medicines organisation may reasonably investigate AI for preparation or review work, yet the same approach can become inappropriate when used to determine a consequential action. NEUVIOR is interested in helping teams define and evaluate that boundary before enthusiasm becomes operational dependence.

Film credits

In the conversation.

NEUVIOR and our founder’s perspectives have appeared across science, business and technology publications.

University of Warwick IATL | AI Ethics NowpharmaphorumTechRadar ProET PharmaFinancial TimesPharmaceutical TechnologyPharmaTimesManufacturing ChemistDigital HealthDaily BusinessTheBusinessDesk.comStartups MagazineInsider MediaYorkshire PostYork PressBusinessCloud

Reporting, commentary, letters and podcast appearances | Read each feature for its format and source.

One label covers very different uses

Drafting a meeting summary, organising published evidence and generating information for a regulatory decision have different consequences. A blanket approval of AI hides those distinctions; a blanket ban may prevent a useful low-risk investigation. A proposed use-case review would describe the task, users, affected people and downstream decisions. It would identify which outputs can remain provisional and which require controls, specialist assessment or a decision that the use case should not proceed.

Use frameworks with their status intact

NIST’s AI Risk Management Framework is voluntary (NIST, 2023). EMA has adopted a reflection paper on AI across the medicinal-product lifecycle (EMA, n.d.). These are useful references, not certificates of compliance or proof of product suitability. Our proposed work would use them as context for a specific evaluation, with current sector requirements considered by the organisation’s qualified advisers. A framework name should never replace evidence that the intended task has been tested appropriately.

Responsibility needs an operational form

A policy saying human oversight is required leaves important questions unanswered. Who checks which part of the output? What evidence can they inspect? Can they reject it without penalty? When is the system unavailable or unsuitable? A useful review makes these responsibilities concrete. It should also examine whether the proposed workflow gives reviewers enough time and information to notice a plausible error, rather than assuming their presence automatically controls the risk.

An adoption decision with a defined boundary

A useful output would state the task that was tested, the circumstances in which it should not be used and the checks needed for any next stage. It would include a named decision owner and unresolved questions. The organisation could then consider a bounded next step without interpreting an encouraging experiment as approval of AI across its business.

A public decision brief | An illustrative example

Open the question.
Inspect the reasoning.

Choose a business question, then look beneath its claim. See what a public source supports, what depends on the local setting and what a useful next step could be.

A medicines or implementation team considering an AI-supported evidence review.

Would a quicker review make the decision better?

The interesting question is the complete piece of work: finding relevant material, checking it, correcting the draft and reaching an accountable decision.

A proposition to investigate

Less preparation. The same ability to challenge.

An evidence-review approach could be useful if it reduces avoidable preparation while preserving source access, relevant qualifications and professional checking. That is the proposed value to test. No time reduction, accuracy improvement or customer result is asserted in this example.

The proposition is constructed for this public brief. It is not a published finding about PHARMORIS.

What the next brief would contain

A decision someone
can commission.

The decision
Whether one defined review task merits an agreed evaluation.
The comparator
The team's existing complete review process, including checking and corrections.
The evidence
Representative cases, source records, reviewer decisions, complete-task effort and exceptions.
Stop or redesign
Stop or redesign if source checking weakens, errors are unacceptable or the complete task offers insufficient value.

Constructed public example | Not PHARMORIS-generated software output or a measured customer result.

Keep this example brief

A bounded proposed PHARMORIS evaluation

The first study could examine one non-live preparation task using approved public or synthetic material. Inputs would include the task definition, expected output, reviewer criteria and examples of known failure conditions. The comparator would be the existing human workflow, including its own errors and effort. Any experimental output would remain separate from operational decisions. The study would not establish a general AI capability or authorise deployment across other tasks.

Measure confidence against performance

Candidate measures include material factual errors, unsupported claims, source traceability, reviewer correction effort and whether users recognise uncertainty. Repeat testing should examine variation between outputs where relevant. The evaluation should count failures that look convincing, not only obvious mistakes. It should also record the cost of supervision. A small time saving can disappear when the checking required for a trustworthy result is included in the workflow.

A stop condition is part of the design

We would stop if the task cannot be separated from live decisions, if confidential material cannot be handled within agreed controls, or if reviewers cannot establish whether an output is supported. The study should pause when a model or source change invalidates the comparison. It should also stop if commercial pressure requires the evaluator to label an unproven use as safe. A credible governance decision can be to narrow the task, change the design or reject adoption.

A useful first conversation

A pharmaceutical executive, digital lead or quality owner can bring one proposed AI task and explain what a user would do with its output. We can discuss the decision boundary, the evidence required and the smallest evaluation that could inform a go-or-no-go choice. The commercial value is an adoption decision grounded in a real workflow and accountable review, with the organisation able to understand what has been tested and what remains unresolved.

Sources and scope

NIST | AI Risk Management Framework | EMA | AI in the medicinal product lifecycle

Discuss a governed AI use case

NIST (2023) AI Risk Management Framework. Available at: https://www.nist.gov/itl/ai-risk-management-framework (Accessed: 14 October 2026).

EMA (n.d.) AI in the medicinal product lifecycle. Available at: https://www.ema.europa.eu/en/scientific-guidelines/use-artificial-intelligence-ai-medicinal-product-lifecycle (Accessed: 14 October 2026).

Another useful angle.

NEWSPerspectives & features
Open newsroom