1 hour ago
30
Anthropic isn’t just building AI models that talk about science anymore. The company has set up physical biology labs and hired bench scientists to let Claude conduct real-world experiments in drug research, a move that blurs the line between AI company and pharmaceutical operation.
The initiative sits under Claude Science, an AI workbench Anthropic launched on June 30, 2026, that integrates over 60 scientific databases and deploys specialized agents for tasks across genomics, proteomics, and computational biology. The platform runs on existing Claude models, including Opus 4.8, and is already being used by pharmaceutical heavyweights like Novo Nordisk and AstraZeneca.
From silicon to petri dishes
What makes this unusual isn’t the computational side. The twist is that Anthropic is pursuing its own internal drug discoveries, specifically targeting rare and neglected diseases, the kind that traditional pharma often ignores because the economics don’t pencil out.
To do this, the company has brought biologists onto its team and built out wet lab spaces where physical experiments can validate what Claude’s models predict. It’s a feedback loop: the AI generates hypotheses and designs molecules computationally, then human scientists test those predictions at the bench, and the results flow back to improve the models.
In August 2026, Anthropic published experiments showing Claude models had designed protein binders against 15 different targets with a hit rate of 22-35%. For context, the typical industry benchmark for protein binder design sits around 10-15%.
Opening the doors wider
On September 17, 2026, the company introduced its Life Sciences Verification Program, designed to expand access to Claude’s biology capabilities across the broader research community.
The program offers grants for both standard and high-risk biology use cases. It also includes a protein design competition run in partnership with Adaptyv Bio, with up to $1M in credits available.
The early adopter list includes Novo Nordisk, the Danish company behind blockbuster GLP-1 drugs, and AstraZeneca, one of Europe’s largest pharmaceutical firms. Various academic labs have also signed on.
Why an AI company is doing wet lab science
The focus on rare and neglected diseases is telling. These conditions affect small patient populations and typically don’t generate the revenue needed to justify traditional pharma R&D investment. If AI can dramatically reduce the cost and time of early-stage drug discovery, diseases that were previously uneconomical to pursue could become viable targets.
Google DeepMind’s AlphaFold transformed protein structure prediction and won a Nobel Prize for the effort. By moving into wet labs and running actual drug programs, Anthropic is staking a claim that Claude can go beyond prediction into the messy, expensive reality of making medicines. Structure prediction tells you what a protein looks like. Designing something that binds to it and works as a drug is a fundamentally harder problem.
The 22-35% hit rate on protein binder design, if it holds up across broader target sets and independent validation, would represent a genuine competitive advantage. A typical new drug costs well over a billion dollars to develop, and most of that expense comes from the high failure rate at every stage of the pipeline.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·