Nearly every biotechnology board has an AI slide now, and most of them are good slides.
Faster target identification. Molecular design that explores a chemical space no team could navigate by hand. Trial planning that stops being a spreadsheet exercise. Regulatory documents drafted in hours rather than weeks. For the commercial case itself, McKinsey’s explainer on AI in life sciences is as useful a summary as any, and there is little in it worth arguing with.
The opportunity is real. AI is already changing how medicines are discovered, developed, manufactured, and monitored.
So the interesting question is no longer whether AI will transform biotechnology. It already has. The more important question is whether an organisation will still be able to explain, justify, and defend the decisions AI influenced, five years after it made them.
That sounds like a compliance question. It is really a question about trust, and trust has always been the foundation this industry rests on.
Biotechnology was a trust industry before it was a technology industry
Seen from outside, this sector appears to run on science. From inside, it runs on confidence.
Patients accept a therapy because they trust it is safe. Physicians prescribe because they trust the evidence. Regulators approve because they trust that the data were generated and reviewed the way the file says they were. Investors commit because they trust that a scientific claim means what it appears to mean.
None of that was an accident, and none of it came from goodwill. It was built, deliberately and expensively, over decades: validation, quality management systems, documentation standards, audit trails, change control, deviation handling, and independent oversight. Read from outside, that list looks like bureaucracy. Read from inside, it is the reason a molecule can travel from a laboratory bench to a patient at all.
Which matters for AI mainly because of where AI is arriving. It is not landing in a governance vacuum. It is landing in one of the few industries that already knows how to build mechanisms for trust, and has the inspection findings to prove what happens when they are missing.
AI has moved from the edge of the work to the centre of the decision
The first wave of AI in life sciences was comfortable. Document summarisation, knowledge search, workflow automation, chat assistants. Useful, occasionally impressive, and almost always at a safe distance from anything that mattered.
The second wave is less comfortable, and it has already arrived. Models now shape target selection, experimental design, deviation prioritisation, clinical reporting, portfolio choices, and quality investigations. In research settings the shift has gone further still: models propose experiments, automated laboratories execute them without human hands, and the results feed back to shape the next cycle at a tempo no manual process matches. As The Conversation has reported, this has outrun both sides of the rulebook — the rules governing biological research do not account for AI-driven automation, and the rules governing AI say nothing specific about its use in biology.
Set the frontier cases aside, though, because the ordinary ones are where most companies are actually exposed. A model that ranks which deviations get investigated first. A model that drafts the clinical study report. A model that flags which batches look unusual.
None of those makes a decision. Every one of them changes one. That distinction is subtle, and it is the whole problem. AI does not need to make decisions to shape them. It only needs to influence the way decisions are reached.
The question nobody puts on the agenda: who owns the decision?
When AI governance reaches an agenda, the agenda is predictable. Bias. Explainability. Hallucination. Data quality. Regulatory compliance. All of it legitimate, and none of it the hardest part.
Underneath sits a plainer problem: decision ownership. Work it through concretely. If a model helps identify a promising therapeutic target, the scientists remain accountable for validating it. If a generative tool drafts a regulatory response, regulatory affairs remains accountable for what it says. If a model influences how a research portfolio is ranked, leadership owns that ranking.
A model may contribute. It may recommend, prioritise, or draft. Accountability cannot be delegated to software.
The hidden governance risk
The realistic danger is not autonomous AI. Nobody is seriously proposing it, and no regulator would accept it.
The realistic danger is diffuse accountability.
As these systems settle into daily work, an organisation gradually loses sight of who relied on a model, how heavily, what evidence was reviewed alongside it, and why the final answer was accepted. Each of those questions has a clear answer on the day the decision is made. Six months later, if nothing recorded them, the answers are gone.
Governance failures of this kind are almost never deliberate. They emerge slowly, from responsibility becoming hard to trace, and they are usually discovered by someone from outside the company asking a question that should have been easy.
Most of what is needed already exists
Here is the part that should be encouraging, and that tends to get lost in the noise about novel AI frameworks.
The disciplines this problem requires are disciplines the sector has practised for thirty years. Risk management. Validation. Documented change control. Quality oversight. Continuous monitoring. Defined accountability.
None of that needs inventing. It needs extending to a new class of system, and to decisions that are now partly shaped by a model rather than wholly by a person. That is a far smaller task than the conference circuit implies, and it has the advantage of being built on machinery the organisation already knows how to operate and already gets inspected on.
What it does require is honesty about scope. Extending a quality system to AI means deciding which decisions are in scope, what evidence each one needs, and who signs. It does not mean writing an AI policy and filing it.
A common objection: this will slow us down
The objection deserves a straight answer, because it is usually made by serious people and it is not stupid.
In a regulated industry, the fear of governance is well earned. Review boards become queues. Validation requirements add four months to six weeks of work. Policies get written so defensively that the only safe action is inaction. Anyone who has run IT in a biotechnology company has seen controls that cost more than the risk they removed.
So it should be said plainly: badly designed governance does slow things down, and a good deal of what is currently sold as AI governance is badly designed. A fifteen-page AI policy that cannot be applied to an actual decision is not a control. It is a liability with a table of contents.
But consider where delay in this sector actually originates. Rarely in the controls applied early, which are cheap. Almost always in what surfaces late: rework, compliance findings, data integrity questions raised two years after a system went live, undocumented changes, inspection observations.
Good governance accelerates scale. Bad governance manufactures bureaucracy. The purpose of the first is not to slow innovation down but to prevent expensive surprises, and the industry’s own history is the evidence. The breakthroughs of the past three decades did not succeed because they were clever; plenty of clever things never reached a patient. They succeeded because regulators, clinicians, payers, and patients trusted the systems standing behind them, and that trust is exactly what allowed them to scale.
Conclusion
The models themselves will become commodities. Many already have. What felt extraordinary two years ago is now a line item available to any competitor with a corporate card, and that trend runs in one direction only.
So a durable advantage will not come from access to AI. It will come from an organisation’s ability to stand behind the decisions AI helps shape.
That is a narrower claim than the usual one about trust, and it is the better one, because it can be tested. It means knowing which decisions are already influenced by a model. It means knowing who owns each of them, what evidence supports them, and how that accountability is maintained once the novelty has worn off.
Technology leadership in biotechnology is quietly becoming governance leadership. That is not what most people in the role signed up for, and it is nonetheless where the work now is.
The organisations that come out ahead will not be the ones that deployed AI fastest. They will be the ones that can explain, without hesitation and without a three-week archaeology project, the decisions AI helped them make.
The regulatory side of this question, i.e. what the EU AI Act actually changes, and why GMP and GCP obligations bite sooner, is the subject of an Akility article: AI regulation in biotechnology: beyond the EU AI Act.
Working with Akility
Akility establishes pragmatic governance across IT, AI, data, quality, and security for Life Sciences and Healthcare organisations, so that innovation delivers measurable value while accountability and human oversight hold. Akility also takes over IT Director or CIO roles, interim, fractional, or long-term. Akility sets the technology vision and leads its delivery, from vision to value.
Articles on this site may be drafted with the assistance of AI tools. Every article is read, checked, and approved by its named author before publication.








