What Might the New Year Herald for Drug Safety, Regulatory and AI Use?
Morgan Waller Scott
Chief of Staff, ArisGlobal
Interviewer: Morgan Waller Scott is Chief of Staff and Head of Marketing Communications at life sciences technology provider ArisGlobal. Her career spans more than two decades, in the orbit of companies and leaders that harness technology, data and design to help drive change in their respective industries.
Abstract:
As 2026 draws to a close, and the life sciences industry looks ahead to a new year and associated strategy revisions, how are efforts to modernise pharmacovigilance and regulatory operations faring, and where do challenges persist? In a recent podcast, ArisGlobal’s Morgan Waller Scott asked three subject specialists for their thoughts.
2026 reflections and 2027 predictions on progress in pharmacovigilance
Morgan Waller Scott (MWS): What would you say have been the defining themes in PV in 2026?
Lucinda Smith (LS): Focus is now moving past isolated pockets of automation (e.g. of intake, data entry, duplicate detection), in recognition that simply automating one step in a fragmented process is not transformational. We have also seen meaningful progress in conversations at a regulator level around AI and its use in PV. The debate is becoming less about whether AI can be used, and more around under what conditions it can be trusted and used effectively. Publications such as the CIOMS XIV report have had a significant impact here.
It is time now to expand the focus into other areas of PV, particularly signal management — with a bearing on how quickly an emerging risk is identified, and how quickly it can be understood and then communicated to patients, preventing harm. The industry also needs to move past human oversight as an assumed final approval step (to have full confidence in AI). Too much time up to now has been spent debating whether AI can or should produce perfect results. The new thinking takes a risk-based model to AI governance, geared to the consequences of a potential error in each use case or scenario.
Morgan Waller Scott (MWS): In a year’s time, what will be materially different about the way PV organisations operate?
Lucinda Smith (LS): I hope we’ll see the value of AI extended to signal management and benefit-risk, with scaled implementation of such applications. This will pave the way for deeper analysis across huge amounts of data from multiple various sources in much less time. Safety scientists will then be able to focus their time on adding valuable human intelligence instead of pulling data together.
That doesn’t mean handing off benefit-risk decisions to an algorithm, but we will see changes in the role of a PV professional and the skills they need, as they’re freed to devote more time to interrogating AI-generated insights, understanding uncertainty, and communicating the implications to patients and regulators and wider organisations. Beyond existing medical, scientific and PV experience, there will be a need for data literacy, for instance.
The year in regulatory operations
Morgan Waller Scott (MWS): What has been the most significant development in Regulatory Affairs this year?
Hans van Leeuwen (HvL): I would say the change in perspective on data governance. Regulatory data was previously seen as a commodity, to be passed from company to regulator. But robust product information is increasingly required cross-functionally within a company, too – for PV, labelling or in a company’s manufacturing or planning systems. Although some larger companies have implemented company-wide schemes for data governance, Regulatory Affairs still generally maintains a key role. In 2026, this has been reflected in data steward roles existing as standard in regulatory departments.
Morgan Waller Scott (MWS): What has been missed if anything?
Hans van Leeuwen (HvL): The concept of collaboration spaces enabling regulators and companies to improve efficiency around submission review has not progressed as expected, because knowledge-sharing processes are proving difficult to implement. The use of AI in health agency interactions will hopefully help with this, by increasing consistency in addressing related questions from different agencies, building the case for information-sharing.
Morgan Waller Scott (MWS): What should Regulatory functions focus on over the next year, in your view?
Hans van Leeuwen (HvL): Now that AI is well proven, regulatory teams should be assessing the scope for associated process redesign. Most pilots today still replicate what teams do today, e.g. to generate documents from CMC or scientific data. Companies then send their dossiers to agencies, which are themselves piloting the use of AI tools to extract data from unstructured text. This is nonsensical. It would be better to focus on the data itself, applied to document generation or other modes of communication such as audio or video, for instance, with a view to reaching patients, healthcare professionals, and users more effectively.
What is ahead for AI in life sciences post-marketing operations?
Morgan Waller Scott (MWS): What was real in AI’s progress in 2026 versus market “noise”?
Jason Bryant (JB): We have seen the gap widen between AI headlines and the underlying reality. On AI safety, credible sources have suggested AI might destroy humanity, yet others dismiss this as doom-mongering. Underneath, though, there is something interesting happening. The frontier labs are slowing or restricting aspects of their own model development because the capabilities have long gone beyond the point where monitoring, alignment and security seriously need to catch up.
On security, headlines talk of models being able to find and exploit vulnerabilities. But trust relies both on discovering vulnerabilities so they can be fixed, as well as engineering for trust by preventing incidents from happening at all. In other words, the discovery of vulnerabilities is not a simple “good” v. “bad” AI story.
Morgan Waller Scott (MWS): What haven’t organisations fully grasped yet about AI?
Jason Bryant (JB): With agentic AI, where trust has to be engineered and evidenced so that the technology can be confidently used in real-world operations, the real enablers now are not accuracy or model capability. Rather, they are everything required for the system to work in a way that people can trust it at scale, in real operations. That includes provenance, “inspectability”, guardrails, lifecycle management, oversight, monitoring, etc.
When delegating some responsibility to agents, it’s important to understand what evidence is being leveraged to reach decisions and trigger actions, and whether that’s sufficient, how consistently it will behave, what happens at the edge/with variation, and the ability to reconstruct what the AI system did. More critically, does the system “know” when human judgement is needed?
Trust needs to be demonstrable. We are seeing some serious work now on how we measure it, at a government level too: the US National Institute of Standards and Technology is looking at evaluation probes inside agentic workflows (inside the “mind” of agentic AI). It is effectively testing whether the evidence is faithful, complete, and adequate.
This is about creating an audit trail around the decisions and the behaviours inside AI models. While “accuracy” was good enough to get AI systems into production, it is proof of trust that will determine how much responsibility we are prepared to give to AI-enabled systems. By the end of 2027, I believe AI leadership in life sciences will be defined by operational maturity.
Morgan Waller Scott (MWS): Any final thoughts?
Jason Bryant (JB): AI is not just about productivity and labour-saving. The bigger question is what new valuable work the industry can afford to do now, thanks to AI, and how it will capture and express that value.
The technology also allows us to consider what happens if the marginal cost of intelligence collapses, for instance; when you can examine the long tail rather than a sample, when monitoring is continuous rather than periodic. These are some of the wider opportunities the industry should start to explore now.
*The full podcast can be found at: https://www.arisglobal.com/podcast/

Lucinda Smith spent 20+ years working in PV at Roche before joining ArisGlobal where she is Chief Safety Product Officer. At Roche, Lucinda led enterprise-wide programmes focused on innovation, transformation and change management. She is passionate about patient safety.

Hans van Leeuwen is a regulatory consultant to ArisGlobal. He previously spent around 25 years in industry, in regulatory operations or IT roles, most recently at Astellas. He previously worked at Merck. Today he draws on his deep industry experience to advocate best practice.
.webp)
Jason Bryant is ArisGlobal’s Senior Vice President for AI Products and the General Manager of NavaX, the company’s AI-first platform for life sciences R&D. Jason is a Data Science Actuary by background, bringing a rigorous decision-science and risk lens to AI product strategy and execution. ArisGlobal is an AI-first life sciences technology company.