The Pistoia Alliance has reported new findings pointing to a growing “scientific content crisis” in life sciences R&D, with many organisations lacking visibility into the data used to train their AI models.

It says the poll, conducted at its US annual conference in Boston, highlights persistent gaps in data quality, licensing processes and AI governance, raising concerns about the reliability of decisions generated by AI tools across research settings.

More than 1 in 4 life sciences professionals (27%) said they do not know what scientific content their organisation’s AI or large language model systems use. A further proportion rely on titles and abstracts alone, limiting transparency over the depth and quality of underlying evidence. Meanwhile, only 36% of respondents said they feed internal documents into their organisation’s models, suggesting many systems are trained on incomplete or weakly traceable datasets.

Speakers at the conference warned that poor visibility over inputs may impair confidence in AI used across discovery, development and clinical trial design. With teams across pharma and biotech evaluating AI to accelerate decision-making, the Alliance noted that data provenance, licensing and model governance must be addressed before tools are deployed at scale.

Neal Dunkinson, senior director at Copyright Clearance Center, said: “It’s clear from discussions at the conference that many AI models are not yet drawing on the full range of scientific evidence needed to deliver authoritative results. Many organizations are still in a learning phase when it comes to both data and governance and, given the stakes for patient safety, that cannot be ignored. Our poll also showed that 38% of respondents say their copyright and licensing policies are unclear or not enforced, meaning many could also be at risk of fines in an already costly drug development process.”

Respondents also identified the lack of shared verification standards as a major barrier to adopting AI agents, with 50% citing this as the biggest challenge. In response, the Alliance is inviting organisations to join its new agentic AI project to help establish common guardrails for safe and scalable deployment. The initiative aims to give R&D teams clearer visibility over how systems learn, adapt and apply data.

Alongside its poll findings, the conference programme featured case studies on the use of AI to support clinical operations and late-stage development. EPAM demonstrated how AI can streamline clinical trial processes, while the Michael J. Fox Foundation outlined how knowledge graphs may help advance research in Parkinson’s disease. AbbVie discussed the technology’s potential for enhancing pharmacovigilance workflows.

Elsevier hosted a roundtable on practical AI adoption, with participants from Eli Lilly, Pfizer, Bayer, J&J and Takeda agreeing that successful deployment depends on problem-led design and integration into existing research workflows. Delegates also highlighted skills shortages as another constraint, echoing Pistoia’s Lab of the Future survey in which 34% reported a lack of specialist talent as a barrier to AI uptake.

Reflecting on discussions at both its US and European meetings, Becky Upton, president of the Pistoia Alliance, said: “It’s notable that the same concerns around AI trust, transparency and skills were raised at both our US and European conferences. These issues are clearly universal across the life sciences community. By working together on common standards, data quality and practical implementation, the life sciences industry will move forward with confidence. The Pistoia Alliance exists to facilitate this collaboration, and we’re excited to carry these discussions into our spring meeting in London.”

The Alliance will hold its spring annual conference at the Royal Society of Medicine in London from 13 to 16 April 2026.