In 2010, launching a Phase III oncology study required approximately 769 steps, 36 approvals, and took a median of 2.5 years.1 Since then, clinical trials have only become more complex, introducing more procedures, endpoints, and data collection requirements2. The incorporation of artificial intelligence (AI) into clinical trials is the next catalyst for more streamlined clinical development.
The immediate future for AI in clinical development is one of partnership with people. Even Microsoft’s AI Diagnostic Orchestrator, (claimed to be four times more accurate than physicians), only achieved such impressive results because it was tested against “raw human performance”, such as physicians working without textbooks, tools, or peer input3. Moreover, recognizing the potential of AI to streamline clinical development, the Duke Clinical Research Institute (DCRI) convened a multidisciplinary think tank to identify areas where AI could offer both immediate benefit and more transformative change4.
A low-risk, high-impact use of generative AI in clinical development is producing customised drafts and documents. AI is rapidly generating content based on existing data, saving time on repetitive tasks while keeping experts in control through review and approval. However, AI tools that interact directly with end users, such as chatbots, introduce additional risks. While they can support sites, patients, and caregivers by answering questions, they are prone to “hallucinations,” where the AI produces confident but incorrect information. This is a known limitation of large language models, with some experts suggesting it may be inherent to how they function.5
Beyond accuracy, privacy is another concern. Generative AI may reproduce fragments of training data, potentially exposing identifiable patient information if datasets are not properly anonymised. While there are mitigation strategies, these risks highlight the need for caution with AI-to-user applications, especially before AI is introduced to more complex areas of clinical development.
Clearly the usage of AI needs governance and oversight to ensure safe and secure application. Even though AI might not get it right all the time, the DCRI think tank concluded that generative AI is likely to offer “acceptable accuracy and relevancy over time6” if the limitations are understood. Even implementing low-risk AI applications, such as tools to automate documentation, support site queries, or streamline study training, will deliver meaningful improvements. For trial sites, which are increasingly strained by the growing number of digital platforms, logins, and administrative requirements, these small, targeted interventions could significantly reduce operational friction7.
So, where are we right now? Although interest in AI is growing, adoption within clinical development remains limited. According to a recent survey of clinical development and pharmaceutical and biotechnology companies, only 10.7% of organizations surveyed had fully implemented AI, while more than a third have yet to adopt it at all8, highlighting that for many organizations, AI remains in the exploratory phase and is considered promising, but not yet embedded into core operations.
As the capabilities of AI evolve, companies that invest in governance, internal skills, and practical knowledge while keeping humans in the loop (HITL) today will be well-placed to use AI effectively and sustainably as its adoption grows. Equally, organizations that delay adoption may find themselves unprepared if employees use AI tools without awareness of the risks or compliance implications.
To ensure safe and responsible implementation in clinical development, AI strategies must be built on ethical safeguards, human oversight, and continuous monitoring. A clear understanding of both the strengths and limitations of AI is essential.
With proper oversight and understanding in place, generative AI can accelerate clinical development in ways we are only beginning to imagine, paving the way for a more effective, inclusive, and patient-centric future9.
3 King, D., Nori, H., The Path to Medical Superintelligence, Microsoft, 30 June (2025)
7 Advarra, New Clinical Trial Industry Survey Reveals Increased Burdens on Sites, October 17 (2023)