Anthropic says six drug makers used Claude in deep partnerships

A chat model that writes code is one thing. A model that reads a drug maker's paperwork and catches the mistakes its scientists missed is another. On 27 October, Anthropic published results from deep research partnerships with six drug makers. The post reached 194 points on Hacker News the same day.

Anthropic named the six companies. They are BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group. Each worked with Anthropic's life sciences team over several months. The post describes a specific task Claude did for four of them, with a measured outcome in each case.

What Claude did for each named partner

At Generate Biomedicines, Claude proposed a genetic design for a receptor used in cell therapy. Anthropic says the design was "about 70% shorter than the standard version". When tested in cells, it let immune cells make "roughly 10 times as many receptors". The quotes are Anthropic's own words.

At Divergence Labs, Claude autonomously generated "roughly 35,000 antibody designs in about a month" against food allergies. Anthropic says 23 of those designs worked: three could bind to the allergy target IgE, and 20 could knock it loose from a receptor.

At Enveda, Claude read manufacturing documents. Anthropic says it "caught all 11 issues that human reviewers had spotted, and flagged seven they'd missed". The company says Claude cut human review time by 89 percent.

At BigHat Biosciences, Claude optimised antibody molecules. The best versions "bound more than 10x more tightly" to their target and clumped "more than 100x more slowly". Every one of these figures is Anthropic's own, from its own write-up of its own work.

What the post does not say

Anthropic does not share the raw data behind these numbers. No independent group has reproduced any of the four results. The post says the work took "several months", so these are not quick demos. It is still one vendor describing one engagement with each partner. The post is marketing in the strict sense: Anthropic chose which numbers to share and which to leave out.

The six partners are also customers. Bristol Myers Squibb and The Column Group get no specific number in the post. Readers get the names but not the task.

AI engineering for drug makers has looked like press releases and vague claims for two years. Named partners with named tasks and measured outcomes is a change in shape. A team selling an AI product into a regulated industry can study each of the four tasks here. Each one picks a job a human reviewer already does. Each one reports how often the model matches or beats the human. That is the shape of a claim another vendor can repeat.

Source

Early progress on accelerating drug development with Claude, Anthropic