The initial reaction across IP-sensitive industries was to slam the door. In early 2024, a ZoomRx survey found that 65% of the top 20 Big Pharmas had banned employees from using ChatGPT, largely relating to concerns that sensitive internal data could be leaked to competitors. That same year, some 83% of life sciences professionals surveyed called the technology “overrated.” Law wasn’t far behind: 78% of US law firms were not using any AI tools at all as of year-end 2024, and more than one in four organizations globally had banned GenAI applications altogether, according to Cisco.
Then adoption exploded. By mid-2025, AI adoption, viewed broadly, was approaching a “tipping point” in the pharma industry, with Define Ventures reporting that the vast majority of pharma leaders planned to increase their AI budgets. Generative AI had “mushroomed from a curiosity to a concern to a now-accepted part of everyday life,” as Pharmaceutical Technology put it in its year-end review. Pfizer built a custom GPT instance called Charlie. Moderna’s CEO described AI tools as “extensions of our team.” In 2025, the FDA itself launched Elsa, a generative AI tool built on Anthropic’s Claude model, after Commissioner Martin Makary directed all FDA centers to begin deploying AI with a target of full integration by June 30, 2025.
Scientists want AI in the Lab… on their terms
Nearly all scientists surveyed want their electronic lab notebooks to evolve into intelligent research partners. But enthusiasm comes with conditions. Eighty-one percent said they would trust AI suggestions only if they can review the underlying science and evidence. That is a bar that generic chatbots don’t clear.
Only 27% said current public generative AI tools meet their scientific needs “very well.” More than half called them useful for some tasks but not others, and 15% said they are poorly suited to scientific work altogether.
Career concerns tracked with seniority. Nearly half (47%) worried that AI would reduce learning and career opportunities for entry-level scientists, while 40% were concerned it could displace demand for senior expertise. Only 17% said they worry AI will take their own job.
Source: Sapio Sciences, “The Rise of the AI Lab Notebook (AILN),” 2026. n=150.
But there’s a gap between the C-suite AI strategy deck and the bench scientist trying to figure out what to do with last Tuesday’s assay data. A new survey from Sapio Sciences found that 45% of scientists in biopharma R&D, CROs, diagnostics and pharma manufacturing are using public generative AI tools through personal accounts to support their work. That figure tracks closely with enterprise-wide data from Netskope, which found 47% of people using generative AI platforms do so through personal accounts their companies aren’t overseeing. Only 5% of scientists surveyed said they could analyze experimental data without specialist support. Sixty-five percent had repeated experiments because prior results were too hard to find or reuse. More than half said their electronic lab notebook was too complex and slowed them down.
Shadow AI use has real risks. IBM’s 2025 Cost of a Data Breach Report found that one in five organizations experienced breaches linked to shadow AI, adding an average of $670,000 in higher breach costs. And in those shadow AI breaches, intellectual property carried the highest cost per record at $178.
The 5% problem
Among the most striking numbers in Sapio’s survey the finding that only 5% of scientists said they can analyze experimental data without specialist support. Put differently, 95% of the scientists generating results in biopharma labs cannot independently act on what they find. They have to wait, for a bioinformatician, a data engineer or another expert.
Image from the “The Rise of the AI Lab Notebook (AILN)” report (reg-walled)
When 65% of scientists say they’ve repeated experiments because prior results were too hard to find or reuse, that’s not a search problem, it’s wasted reagents, burned instrument time and delayed programs. Sapio’s survey found that only 7% of scientists can configure their ELN for new assays or workflows without help, and 71% say the tools are hard to configure or adapt at all. In pharmaceutical manufacturing environments, that configuration frustration runs to 84%.
Across the full sample, 51% of scientists said they spend too much time importing and exporting data, but among U.S.-based scientists, that number jumps to 81%. In pharma manufacturing, it’s 72%.
Independent data tells much the same story
A separate Benchling survey of more than 600 scientists found that researchers in legacy lab environments spend nearly half their week on basic data entry and process-related tasks. Those scientists reported only 60% confidence in data accuracy and described 3 to 5 handoffs every time data moves between teams.
A blinded survey of 300 biopharma R&D and IT professionals commissioned by Benchling found that nearly half of organizations reported fewer than three out of five lab instruments connected to software that supports data capture. The top barriers cited: no standard instrument data formats and not enough internal resources to write integrations. Scientists are still walking USB drives between machines.
What scientists say they want
When Sapio asked what the next generation of ELNs should look like, the answers had much to do about removing the dependency on specialist support that the 5% figure exposes. Ninety-six percent said future ELNs should help interpret data. Ninety-five percent wanted conversational, text-based interfaces. In other words, they want the kind of interaction they’re already getting from personal ChatGPT, Claude or Gemini accounts, just without the IP risk.
Demand for built-in scientific AI capabilities varied by discipline: 83% of diagnostics labs and 74% of biopharma R&D wanted retrosynthesis, toxicity, and solubility prediction; 71% of biopharma R&D wanted molecular binding simulations; and 65% of CROs wanted genetic sequence optimization.
“Most ELNs were designed as tools that focused on documenting experiments, not actively supporting scientists or guiding next steps,” said Mike Hampton, chief commercial officer at Sapio Sciences. “Today, scientists are working with increasingly complex data and are expected to move from results to decisions faster than ever, yet many ELNs still function like glorified filing cabinets.”
