As AI drives seismic change across the pharmaceutical industry, its adoption is accompanied by uncertainty and risk. Questions around accuracy, data security and accountability are already familiar: Can AI outputs be trusted? Is personal data protected? Are decisions being made by humans or machines? Within healthcare and pharma, where decisions can have profound and lasting consequences, these concerns are particularly acute.

At Sprout Health Solutions, the patient voice is central to everything we do. We have welcomed the steady evolution of patient engagement within pharma from ‘nice to have’ to regulatory and scientific expectation. Increasingly, patient input is recognised not only as ethically important, but as critical to research quality, relevance and credibility.

This shift has driven earlier and more meaningful adoption of patient-centric insights, with clear downstream benefits. We are strong advocates of patient involvement at all stages of the drug development pathway, from preclinical research and early discovery through to clinical trials and post-marketing deployment.

Despite the transformative potential of novel AI technologies, the fundamental principles of patient-centred research remain unchanged. In fact, these principles are more important than ever as AI intensifies the trade-offs between speed and efficiency and the need for accuracy, oversight and patient trust.

In practice, this means continuing to champion patient engagement from the outset and maintaining a consistent focus on patient perspectives, needs and outcomes. As organisations race to accelerate AI adoption, it is imperative that the core principles of patient-centricity are not lost in the process.

In our work with patients and caregivers across a wide range of rare diseases, one challenge we hear repeatedly relates to the scarcity of reliable, accessible information. Many people living with rare conditions must navigate fragmented evidence, inconsistent guidance and prolonged periods of uncertainty.

In this context, generative AI offers a compelling opportunity. The rapid collation and synthesis of information could democratise access to knowledge, support self-advocacy and enable more informed decision-making for patients and their families. Used thoughtfully, these capabilities could help to address long-standing unmet information needs within rare disease communities.

However, these benefits must be weighed against important limitations. Generative AI is only as robust as the data on which it is trained, and rare disease data sets are often small, heterogeneous and shaped by historical inequities. AI models built on such foundations risk entrenching existing biases, overlooking underrepresented experiences, or producing outputs that appear authoritative but lack real-world relevance. Moreover, the potential for so-called ‘hallucinations’, where inaccurate or misleading information is presented with confidence, further heightens the risk, particularly in healthcare settings where trust and accuracy are paramount.

This tension highlights the complex trade-offs inherent to AI adoption. While AI may reshape the landscape, it does not change who that evidence is ultimately for. Patients must therefore be afforded a meaningful role in shaping how AI tools are designed, developed and applied. Only they can weigh the benefits of efficiency and access against the risks of inaccuracy or misrepresentation from the perspective that truly matters.

We therefore advocate for a ‘patient-in-the-loop’ approach that places lived experience at the heart of AI-enabled healthcare innovation. When we talk to patients and caregivers, we find a typically varied picture. While many recognise the potential utility of AI, a meaningful minority find its use in healthcare concerning, reflecting deep-seated worries around accuracy, data privacy and the erosion of human oversight.

This serves as a timely reminder that the patient perspective should never be taken for granted. As AI continues to reshape the pharmaceutical landscape, efficiency and innovation must be carefully balanced against transparency, inclusivity and accountability, particularly in areas such as rare disease, where patient voices have historically been underrepresented.

Patients should never be treated as passive recipients of AI-enabled outputs. The enduring principles of patient-centred research show that it is precisely the active engagement of patients in the design, development and governance of AI that will unlock its potential, aligned to the needs, values and expectations of those it ultimately serves.