Anthropic Moves From AI Research Tools Into AI Drug Discovery

Two announcements, one strategy

On June 30, at an event in San Francisco, Anthropic stopped positioning itself as a vendor to AI drug discovery and started acting like a participant in it. The company unveiled Claude Science, an AI workbench built for biopharmaceutical researchers, and in the same breath announced it would run its own internal drug discovery program targeting diseases that traditional pharmaceutical companies consider commercially unattractive.

The pairing is the story. Plenty of AI labs sell tools to scientists. Fewer AI labs turn around and use those tools to try to find drugs themselves, before selling the tooling that made it possible.

Anthropic's life sciences head, Eric Kauderer-Abrams, was direct about why the company is doing both at once. Building the right models and products for biopharma, he said, requires living the work alongside the customers Anthropic is trying to sell to. Jonah Cool, Anthropic's head of life sciences partnerships, framed the neglected disease focus as central to that dual mission: build and sell AI tools for life sciences companies while running programs the commercial market has no incentive to fund.

Why a company that sells AI needs to develop drugs too

Anthropic has not said what it will do if its own drug discovery program produces a promising candidate. Kauderer-Abrams did not specify a path to clinical trials, the stage where a traditional biopharma company would normally take over. An Anthropic spokesperson told CNBC that as a public benefit corporation, the company can select programs based on patient benefit rather than commercial return, including work the broader market ignores.

That structure matters for how to read the move. A public benefit corporation has more room to run a loss-making research arm than a standard for-profit AI company would. It also gives Anthropic a credible answer to a question every AI drug discovery vendor eventually faces from biopharma buyers: has your own tool ever actually found anything? Running an internal pipeline, even a narrow one focused on neglected diseases, is a way to generate a real answer rather than a marketing claim.

Anthropic backed the pivot with more than a product launch and a research program. In April 2026, it paid roughly $400 million in stock for Coefficient Bio, an eight-person startup founded by former Genentech computational biologists building AI tooling for drug-pipeline planning, target identification, molecular design, and clinical trial strategy. That acquisition gave Anthropic a life sciences team with domain experience the company did not have in-house before this year.

The AI drug discovery companies Anthropic just picked a fight with

Anthropic is not entering an open field. Google DeepMind's Isomorphic Labs has standing partnerships with Eli Lilly and Novartis worth billions in milestone payments, raised a $2.1 billion Series B in May 2026, and is already closer to clinical entry with AI-designed drug candidates than Anthropic's newly announced program. OpenAI launched GPT-Rosalind in April, a reasoning model built specifically for biology and translational medicine, available through a restricted Trusted Access Program to enterprise customers including Novo Nordisk, Amgen, Moderna, and the Allen Institute.

The three approaches are not identical. Isomorphic sells outcomes: it runs its own drug-design engine and hands pharma partners actual candidates under commercial terms. OpenAI sells access to a reasoning layer that accelerates literature review, experiment planning, and evidence synthesis without claiming to design molecules on its own. Anthropic's Claude Science sits closer to the OpenAI model structurally, a workbench built on existing Claude models rather than a new purpose-built one, but the addition of an internal drug program pushes Anthropic toward the Isomorphic side of the spectrum too.

For US enterprise AI buyers evaluating vendors in this category, the practical distinction is what each company is actually optimizing for: Isomorphic wants clinical candidates, OpenAI wants adoption of a reasoning model inside existing pipelines, and Anthropic is now trying to do a version of both while using Novo Nordisk and the Allen Institute as its own early customer references, the same two organizations OpenAI cited at the GPT-Rosalind launch.

What actually ships inside Claude Science

Claude Science is not a new model. It is a research environment built on top of Anthropic's existing Claude models, integrating more than 60 scientific databases across genomics, proteomics, and cheminformatics into a single workbench. It runs on NVIDIA's Agent Toolkit, connecting to life sciences models available through NVIDIA BioNeMo, including Evo 2, Boltz-2, and OpenFold3, rather than replacing them.

Every artifact the platform generates comes with its underlying code, computing environment, a plain-language explanation of the process, and a complete conversation history, a design choice aimed at reproducibility in a field where regulators and internal review boards need to trace how a result was produced. The application runs locally on macOS or Linux, or through a remote machine, and is available in beta to Pro, Max, Team, and Enterprise Claude subscribers.

Anthropic is also offering up to $30,000 in compute credits across as many as 50 research projects to encourage early adoption. Applications close July 15, 2026, with recipients announced by July 31. Early customers named at launch include Novo Nordisk and the Allen Institute, where neuroscientist Jerome Lecoq built multi-agent workflows inside Claude Science to extract claims and quantitative findings from thousands of papers, and UCSF's Brain Tumor Center, which used the tool to cut the time needed for germline analysis of gliomas.

The question Anthropic did not answer at launch

Kauderer-Abrams did not say what happens if Anthropic's internal program identifies a promising drug candidate. That is normally the point where a biopharma company begins clinical trials, a process requiring regulatory expertise, manufacturing capacity, and capital that an AI company does not currently have in-house. Whether Anthropic intends to license candidates to established pharma partners, spin up its own clinical operation, or simply use the exercise for internal learning remains unstated.

The market reaction to the announcement suggests investors are taking the entry seriously regardless. Shares of Schrodinger, a computational drug discovery company, fell as much as 8.3% intraday following the announcement, while Recursion Pharmaceuticals saw volatile trading before partially recovering. Both companies compete in adjacent territory to what Anthropic just described.

The timing also lands inside a broader Anthropic story. The move follows the brief suspension of Claude Fable 5 and Claude Mythos 5 in mid-June under US export controls, restored July 1, and comes amid reporting that both Anthropic and OpenAI may be moving toward IPOs. A B2B expansion into a large, high-margin vertical like biopharma is the kind of revenue diversification a company preparing for public markets would want on the record before filing.

What to watch

• Whether Anthropic names a clinical trial partner or path for any candidate its internal program produces.
• How OpenAI and Google DeepMind respond, given both already have named biopharma customers overlapping with Anthropic's early list.
• Uptake on the $30,000 credit program before the July 15 application deadline.
• Whether Claude Science pricing and access expand beyond the current Pro, Max, Team, and Enterprise beta once out of beta.

Sources