Elsevier has launched LeapSpace, a new AI-powered tool designed to help researchers analyze scientific literature by scanning millions of paywalled journal articles. The tool, built on a large language model, allows users to ask complex questions—such as whether existing drugs could be repurposed for Parkinson’s disease—and returns answers with cited sources from the original research. What sets LeapSpace apart is its unprecedented access to full-text content from Elsevier and four major publishing partners: Emerald, the Institute of Physics, the New England Journal of Medicine Group, and Sage Publications. Together, their collections include 18 million articles and books, making it one of the largest curated databases of paywalled scientific content available for AI analysis. The tool is not free. Users must pay either through an institutional subscription—based on the size and research output of the institution—or a personal plan costing $32 per month. Access to the cited papers themselves still requires separate subscriptions, creating a layered cost structure similar to paying for multiple streaming services. This model has drawn criticism from experts who argue that it fragments access to knowledge and limits the tool’s usefulness. Jason Priem, CEO of OpenAlex, a free bibliographic database, said the approach of isolating and monetizing portions of the literature fails to serve the broader research ecosystem. “You can’t understand science by only seeing a piece of it,” he said. While other AI tools like Consensus and Asta rely mostly on open-access articles due to limited access to paywalled content, LeapSpace’s strength lies in its access to a vast, proprietary corpus. However, this advantage comes with trade-offs. Researchers using general-purpose models like ChatGPT often have no idea how much paywalled content those systems have been trained on. LeapSpace, by contrast, promises transparency: it will not favor its own publisher’s content, user queries will remain private, and the underlying LLMs are not trained on user data. LeapSpace includes features designed to improve trust and usability. One such feature is the “Trust Card,” which explains why a particular citation was included, helping users assess the reliability of the response. The tool also identifies funding opportunities and potential collaborators, adding practical value beyond simple summarization. Still, concerns remain. Experts note that while LeapSpace covers about 22% of all research articles published in 2024, nearly half of those are behind paywalls. Meanwhile, about half of all papers published that year were open access, potentially available to other AI tools—though some licenses restrict AI use. Elsevier says it plans to expand its open-access inclusion as licensing allows. Critics also warn of growing market concentration. Dave Hansen of the Authors Alliance expressed concern that the coordinated effort among major publishers to create a unified AI product could entrench Elsevier’s dominance. “Elsevier is already huge,” he said, “and this kind of collaboration could limit competition and innovation.” There is also no standardized way to evaluate the accuracy of AI-generated summaries. Jevin West, an information scientist at the University of Washington, noted that LLMs are skilled at producing plausible-sounding answers, making it hard for users to distinguish between correct and misleading information. “They’re good at pleasing us, but not necessarily at being right.” Despite these concerns, Elsevier reports strong early feedback from institutions adopting the tool, with executives highlighting significant time savings and productivity gains. Erik Engstrom, CEO of RELX Group, Elsevier’s parent company, emphasized that the company plans to limit content licensing to AI developers, viewing direct access to its data as a core competitive advantage. As the volume of scientific literature continues to grow, tools like LeapSpace may become essential for researchers navigating complex, interdisciplinary fields. But whether their high cost and restricted access justify their value remains a key question in the evolving landscape of AI and scholarly communication.