The short answer, for most organisations, is: not really. Not yet.

AI is already inside pharmacovigilance and medical information workflows. Case intake tools, literature screening software, enquiry response drafting, signal detection platforms. It didn’t arrive with fanfare. It arrived quietly, embedded in systems people use every day. And in many cases, the professionals using those systems have received little to no structured guidance on what AI actually is, how it fails, or when not to trust it.

The Problem Isn’t Resistance. It’s the Absence of Structure. There’s a tendency to frame AI adoption as a culture problem. Employees are hesitant. They don’t trust the tools. They need to be convinced.

That framing misses the real issue. Most people aren’t resisting AI out of stubbornness. They’re uncertain because nobody has told them what “good” looks like. What should a pharmacovigilance associate actually be able to do with AI? What should a medical information manager be responsible for? What does it mean to use these tools safely in a regulated environment?

Without clear answers, you get two failure modes: people who avoid AI entirely, and people who use it without the critical eye it requires. Neither is acceptable in functions where errors have patient safety consequences.

The risks of AI misuse in case narrative drafting are not identical to the risks in generating responses to HCP enquiries. Any structured approach to AI literacy needs to reflect that distinction.

Five Levels, Not One Training Session

AI literacy in pharmacovigilance and medical information is not a box to tick. It develops in stages, and organisations need to think about it that way.

At the foundation, every person in these functions needs a basic working understanding of what AI is, where it already appears in their workflows, why data quality matters, and why human review is not optional. For pharmacovigilance professionals, that means recognising that AI may already be flagging potential adverse events in their inbox. For medical information professionals, it means knowing that AI-generated content still needs to be verified before anything reaches a healthcare professional.

From there, people progress to actually using tools, with appropriate checks in place. The critical competencies at this stage are data privacy (what can you put into a publicly available AI tool? less than most people assume), bias awareness, and the ability to sense-check an output rather than accept it at face value.

The next stage is genuine integration into daily work, not occasional use when convenient. At this point, professionals are comparing AI outputs against source data, flagging inconsistencies, and actively participating in decisions about responsible deployment.

Beyond that sits the evaluator and mentor level. These are the people who understand how a model generates its answers, not at a developer level, but well enough to recognise when something might be wrong and why. They can identify hallucinations, spot when a tool is underperforming for a specific task, support less experienced colleagues, and contribute to validation activities. In pharmacovigilance, that might mean reviewing AI-assisted signal detection for reliability. In medical information, it might mean overseeing the governance of AI-generated content for accuracy and compliance.

At the top sits leadership: defining AI policy, setting governance structures, representing the function externally. Most organisations will have only a handful of people operating here, but they’re the ones who determine whether AI gets embedded responsibly or recklessly.

What Behaviour Actually Looks Like at Each Level?

Abstract descriptions of competency are only useful up to a point. What matters is being able to recognise what different levels of AI literacy actually look like in practice.

At the unfamiliar end, a pharmacovigilance professional dismisses AI as unreliable and relies entirely on manual processes. A medical information professional avoids AI in enquiry triage and treats any AI output as suspect by default. These aren’t hypothetical caricatures. They’re real behaviours in real teams.

At the strategic level, a pharmacovigilance leader is building AI-enabled safety analytics and developing no-code agents. A medical information leader is shaping the enterprise Medical Information AI strategy and overseeing governance for patient-facing responses.

The gap between those two points is significant. Making the intermediate steps visible and achievable is how organisations actually move people along that path.

Embedding AI Tools Is Harder Than Selecting Them

Something that gets overlooked in AI adoption discussions is the practical difficulty of actually embedding a tool into a workflow. A few things are worth flagging.

On data security: where does the tool store your data? If you’re using a publicly available AI tool and uploading clinical trial efficacy data, that data becomes accessible to every other user of that platform. This isn’t a hypothetical risk. It’s a real one that organisations routinely underestimate when moving quickly to adopt new technology.

On validation: every AI tool needs rigorous testing before it goes live, and a monitoring plan for after. That means checking training datasets, identifying biases, and establishing clear escalation pathways for when outputs fall outside expected parameters. It also means maintaining an issue log and having a documented process for serious problems, including the option to switch the tool off while you investigate.

On governance: who is responsible for AI oversight in your pharmacovigilance or medical information function? If you can’t answer that question immediately, that gap needs closing before the next tool rolls out.

The Regulatory Clock Is Running

The EU AI Act (Regulation (EU) 2024/1689) is the first comprehensive legal framework for AI anywhere in the world. It won’t be the last. AI regulations vary by region and will continue to evolve. Organisations that treat AI literacy as a one-time training exercise rather than an ongoing professional capability are going to find themselves scrambling to catch up.

What constitutes an appropriate level of AI literacy for a pharmacovigilance professional will look different in three years than it does today. Building structures that can be updated and refined over time is not optional. It’s the only approach that works.

The Bottom Line

Pharmacovigilance and medical information functions operate under tight regulatory scrutiny, with direct consequences for patient safety when things go wrong. AI introduces genuine efficiency gains in these areas: faster literature screening, better signal detection, more consistent case documentation. It also introduces new failure modes. Hallucinated references. Misclassified seriousness. Outputs that look authoritative and are subtly wrong.

The organisations that will navigate this well are the ones that treat AI literacy as a professional capability like any other: something that can be assessed, developed, and improved over time, with clear expectations at every level. The work of building those capabilities is something each organisation has to do for itself.