Feature · Anthropic drug development

35,000 antibodies in a month, document review cut from 25 hours to 3 — six case studies from Anthropic's push into drug development

At biotech Enveda, chemists spent up to 75% of their time on paperwork. Fed 1,061 pages of manufacturing documents, Claude caught all 11 issues human reviewers found and seven more. Anthropic's Oct. 27 post has six more cases. Here is where Claude went in, what changed and what is still unknown.

1. The bottleneck isn't one place: six partners, months of work, the whole pipeline

   Anthropic's life-sciences team worked with BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group. Bottlenecks are spread from target selection to molecule design, manufacturing, trial paperwork and trial design, it said, so the way to find them is to get deep into real programs.

2. Generate: a design 70% shorter produced about 10x the receptors

   Generate wants CAR T therapy to work on solid tumors. Six scientists used Claude Code to propose designs for one genetic part. A design about 70% shorter than standard did better in lab tests: roughly 10x as many cancer-spotting receptors, more efficient tumor killing and longer-lasting immune cells.

3. Divergence: 35,000 antibody designs in a month, three that beat the standard drug

   Divergence is building a food-allergy drug that detaches IgE from immune cells. Largely autonomously, Claude generated about 35,000 designs in a month. Early tests found 3 that bind IgE and 20 that dislodge it; 3 of those 20 dislodged more than the standard medication.

4. BigHat: 1,010 variants in 10 weeks; best versions bound 10x tighter

   Claude proposed changes and BigHat's lab tested about 1,010 variants in about 10 weeks. After doubting the standard stability metrics, Claude and a researcher reanalyzed the data and proposed new ones. The best versions bound over 10x tighter, one stayed in solution 18°C higher, and another clumped over 100x more slowly.

5. Enveda: 75% of chemists' time on paperwork; review down 89%

   On 1,061 pages Claude caught all 11 issues humans found and flagged seven they missed, cutting review from over 25 hours to under three. Claude drafted one key regulatory document in four hours; by hand it takes about three months.

6. The Column Group and BMS: can AI make riskier ideas investable?

   The Column Group is founding Pleiades to run many more drug programs with a small team; its first Claude skill was evaluated on 45 target-disease pairs. BMS uses Claude to assess early-trial plans and estimate success odds, with the work set to expand in 2027.

7. Mithril Medicine: one person, 11 labs, mouse results in under 18 months

   Anthropic acquired the one-person company; founder Cody Tranbarger ran contracts, logistics and experiments through Claude. Human trials could start as soon as late 2028 if the candidates hold up. Polymarket's post on it drew 1,700 likes.

8. What's still unknown: mostly preclinical, and the results are the company's own

   Anthropic calls the work early and mostly preclinical. Partner figures come from technical reports and preprints; we found no independent verification at the time of writing. Anthropic plans to open its models wider through Claude Science, credits for life scientists and a Life Sciences Verification Program.