This month, CAS, a division of the American Chemical Society specializing in scientific knowledge management, announced the launch of CAS Newton, a conversational agentic artificial intelligence designed for scientific investigation. Newton is rooted in the organization’s highly structured collections of literature and data, spanning chemistry, biology, materials science, and intellectual property.
“We don’t look at Newton as just pure AI, right? Newton is a marriage of AI and peer-reviewed, proven science. And only CAS is uniquely qualified to create that marriage,” says Jian Wu, interim vice president of product management at CAS, citing its long history of curating scientific information.
Newton can be used within SciFinder and BioFinder, in addition to use as a standalone interface. And CAS is working on secure integrations for organizations to use the tool alongside their proprietary data, a development that will be announced soon, Wu says.
Among other scientific capabilities, Newton can perform spectra prediction, structure search, and functional-group reactivity prediction. The AI uses a multi-agent framework to pull together the content and tools needed for the subject matter at hand, says Andrea Jacobs, CAS director of data analytics.
One application of Newton is drug discovery. A user could inquire about the chemical reactions needed to optimize a lead compound within a specific scaffold and having metabolic issues in part of its structure, Jacobs says. “Being able to ask that type of question and not just have that be like, here’s 72 search-result sets that you’re going to have to sift through and think about which parts actually matter, but actually get a response that’s directly answering that sophisticated multipart question—it’s transformative.”
According to CAS, user inquiries won’t be used for AI training, the infrastructure is private, and all output can be audited. Still, users of any AI would need to verify its conclusions. “It’s really dangerous for a scientist to live in a space where an answer looks like it might be right but isn’t grounded in trustworthy scientific information,” Jacobs says, adding that “specialist AI like CAS Newton really can connect the dots in a way that nonspecialist tools can’t.”
To avoid hallucinations, CAS limits Newton to working from curated sets of knowledge and performs post-evaluations to assess its output, she says. “Of course, AI needs to be used responsibly. You know, scientists’ jobs don’t go away by virtue of AI being good at connecting the dots in certain aspects of science.”
2026 American Chemical Society
