The Pistoia Alliance has released data indicating a growing “scientific content crisis” that is limiting the accuracy and adoption of AI in R&D. The poll found that more than 1 in 4 life science professionals (27%) do not know what scientific content their organization’s AI or LLM systems use or rely only on titles and abstracts.
Meanwhile, only around 1 in 3 (36%) are plugging internal documents into models. As a result, many AI systems are being built on incomplete or insufficiently traceable scientific evidence, reducing confidence in the reliability of AI outputs. The poll was conducted at the Pistoia Alliance’s annual US conference held in Boston, which convened more than 170 experts from pharma, technology and academia to explore AI challenges and the guardrails needed for its safe use.
“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,” said Neal Dunkinson, Senior Director, CCC (Copyright Clearance Center). “To ensure models are grounded in the highest-quality and most complete scientific datasets, the industry must ensure any datasets being used are AI-ready: meaning properly structured, licensed and transparent.”
Another theme that emerged from the poll findings was the need for stronger benchmarking and governance for AI agents to give organizations full visibility over which data models are learning from. The Alliance’s poll data reinforced this, with half (50%) of respondents identifying the lack of shared verification standards as the biggest barrier to agent adoption.
Read Pistoia Alliance's announcement here
