← All posts
This post is written in English.
Anthropic published a progress report on 27 October covering drug development work with partners. It's worth reading with one sentence in mind, from the post itself: "All of this research is still at an early stage."
What's reported
- Generate Biomedicines: a CAR-T genetic component about 70% shorter than standard which, in lab tests, let immune cells make roughly 10 times as many receptors.
- Divergence Labs: about 35,000 antibody designs in a month for food allergies, with 3 of 20 hits dislodging IgE more than the standard medication.
- BigHat Biosciences: about 1,010 variants tested over 10 weeks, with the best binding more than 10 times more tightly.
- Enveda: on 1,061 pages of documents, Claude caught all 11 issues human reviewers found and flagged seven they missed, cutting review time by 89%.
- Mithril: a possible rare muscular dystrophy treatment reached promising mouse results in under 18 months. Anthropic has acquired this company, and a human trial could come as soon as late 2028 if candidates hold up.
How to read this
These are lab results, not clinical outcomes, and most drugs that look good at this stage fail later. The post doesn't say whether any results are peer reviewed. They're also published by the company selling the model, about partners using it, and Anthropic now owns one of the programmes.
It's also hard to know how much of each result came from the AI, compared with skilled scientists using a good tool.
My read
The document review example may be the most immediately real, even though it's the least exciting. Regulatory paperwork is slow and expensive, and speeding it up doesn't need a scientific breakthrough. It still needs people checking it, which is what the post describes.
For the science, the test is human trials, and those take years. I'd come back to this in 2030.
