Drugmakers say AI is cutting discovery times, and they're watching Anthropic
GSK, Novartis, Sanofi and others say AI has cut early drug development time by as much as half. They're also sizing up Anthropic, now a drug developer itself.
Several of the world's largest drugmakers say AI has cut the time it takes to produce new molecules and move them into the clinic by as much as half this year. Reuters reported on Nov. 20 that the same companies are now sizing up Anthropic, which has moved from supplying AI tools to running drug programs of its own.
That second part is the news for Anthropic. Pharma executives can't agree on whether it's a vendor, a partner or a competitor.
Anthropic's move into drug programs
Anthropic's own research update, published Oct. 27, lays out the shift. The company's life sciences team has worked with six organizations: BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group.
It has also started its own programs, including ones for rare childhood forms of muscular dystrophy, where Claude handles much of the scientific and operational work.
As part of that, Anthropic acquired Mithril Medicine, a one-person company developing treatments for two rare forms of muscular dystrophy. Anthropic says Mithril went from an idea to promising results in mice in under 18 months. If the drug candidates keep holding up, the company expects a drug could be ready to start human clinical trials as soon as late 2028.
Anthropic is blunt that this is early. It describes the work as mostly preclinical, meaning not yet at the stage of human testing.
There's a software side too. Anthropic launched a research-focused product called Claude Science and added Novartis CEO Vas Narasimhan to its board. Other companies, including Novo Nordisk, use Anthropic models and Claude Science in their R&D work.
The numbers drugmakers are citing
GSK, Bristol Myers Squibb, Novartis, Eli Lilly, Novo Nordisk, AstraZeneca and Sanofi all told Reuters they've moved past paperwork tasks such as regulatory filings. They now use AI to design molecules, rank the promising ones and flag safety risks. AstraZeneca's Sharon Barr said there's no part of the discovery-to-development pipeline where the company isn't using AI.
- Novo Nordisk: AI has cut the design, manufacture and test cycles for new molecules from around eight to three, and each cycle from eight weeks to four.
- Novartis: Fiona Marshall, its president of biomedical research, said the average time from project start to a drug candidate has fallen from three years to about 18 months to two years.
- Sanofi: The company said an AI-enabled approach saved at least a year on asthma treatment lunsekimig by letting it skip a dose-selection study. Matt Truppo, its head of computational and AI strategy for R&D, expects early-discovery timelines to fall 20% to 25% within two years, and perhaps by half within five years, across more than 100 programs.
These are company claims, not independent audits. I'd probably treat the headline "half" as a best case.
The limits are real
The pipeline of AI-enabled drugs is growing slowly. Researchers identified 129 AI drug candidates across 69 companies in March, up from around 117 eight months earlier, and most are still in early development.
Dr. Waqas Haque, an oncology fellow at University of Chicago who worked on that count, expects a sharp rise in AI-developed drugs over the next few years. He said he doesn't think they'll be approved any quicker, because human trials and regulators remain bottlenecks.
Drug development is slow and risky without AI. The industry lobby group PhRMA says it can take 10 to 15 years and cost billions of dollars to bring a drug to market, and nearly 9 in 10 conventionally developed drugs fail in clinical trials.
Aliza Apple, who runs Lilly's TuneLab, pushed back on claims from AI leaders that the technology could cure all disease. "It would be great for them to cure one, and then let's talk about the rest," she said.
Partner, vendor or rival
Views differ by company. GSK's chief scientific officer, Tony Wood, said he sees Anthropic "as another competitor". Bristol Myers and Novo see AI companies more as partners.
Apple said Lilly once saw such firms as vendors, which is how they framed themselves, and now increasingly sees them as collaborators. Her take on the future was short: a year from now is anyone's guess.
Data is the sticking point. Novo and Sanofi both said they don't give proprietary data to frontier labs for model training. Truppo's reasoning was that a model breaks down quickly without the data to train it on.
Luca Giani, a senior principal at Alumni Ventures, said Anthropic and rival frontier labs are shopping for pharma-related acquisitions. Anthropic didn't respond to Reuters' request for comment on its acquisition plans or its relationship with drugmakers.
Next steps
Anthropic says it plans to scale its drug work up substantially. It's also taking proposals from outside research groups through an open call for guest investigators in the life sciences.
Bristol Myers Squibb is using Claude to assess early-stage clinical trial plans and estimate their odds of success. Anthropic says the two plan to expand that work in 2027.
The clearest milestone is Mithril's candidate, which Anthropic says won't reach human trials before late 2028 at the earliest.
Honestly, I think the awkward part for drugmakers is buying from a supplier that also runs a pipeline. As far as I can tell, no one has shown yet that an AI-driven program produces a better drug, only that parts of the process run faster. I'd wait for clinical data before reading much into either side's timelines.