Anthropic says Claude designed 35,000 antibodies, cut filing reviews 89%

Summary

• Anthropic published results from months-long drug research partnerships with six developers plus its own discovery programs.
• Claude generated antibody designs, re-analyzed lab data and drafted regulatory documents across the partnerships.
• The numbers point to smaller teams running more drug programs than headcount normally allows.

The latest

Claude independently generated roughly 35,000 antibody designs in about a month for a food-allergy treatment and flagged every issue human reviewers caught across 1,061 pages of manufacturing records, Anthropic said on 27 October 2026. Its life sciences team published results from partnerships with six drug developers, alongside in-house programs targeting rare childhood muscular dystrophies.

Details

• The partners: Anthropic said the work covered BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group. The company framed the goal as mapping both the strength and the limits of its models across drug development tasks, and as producing case studies for the life sciences community.
• The acquisition: Anthropic acquired Mithril Medicine, a one-person company developing treatments for two rare muscular dystrophies. Founder Cody Tranbarger used Claude to manage contracts, logistics and experiments with 11 external labs. One candidate moved from idea to promising mouse results in under 18 months. The deal value was not disclosed.
• Cancer cell engineering: With Generate Biomedicines, six scientists used Claude Code to propose designs aimed at making CAR T therapy work against solid tumors. One design, about 70% shorter than the standard version, let immune cells produce roughly 10 times more receptors, kill cancer cells more efficiently and stay active longer.
• The allergy program: Divergence Labs is developing a food-allergy treatment that detaches IgE from immune cells rather than only neutralizing circulating IgE, a design requiring two arms. Early testing found 3 designs that bind IgE and 20 that detach it, with 3 of the 20 outperforming the standard drug. Combining both arms is next.
• The lab reversal: BigHat's lab made and tested about 1,010 variants in roughly 10 weeks. Midway, the team suspected standard thermal stability and aggregation metrics were misleading; Claude and an Anthropic researcher re-analyzed raw data and proposed new metrics and a protocol change that BigHat tested. The best versions bound more than 10 times more tightly, Anthropic said.
• The BigHat numbers: The cancer-targeting candidate stayed soluble at temperatures 18 degrees Celsius higher, and the long-acting candidate aggregated more than 100 times slower.
• The paperwork: Enveda's manufacturing chemists had been spending up to 75% of their time on paperwork. Across 1,061 pages, Claude caught all 11 issues human reviewers found plus 7 they missed, and human review time fell 89%, from over 25 hours to under 3.
• Regulatory drafting: Claude wrote first drafts of two regulatory documents for Enveda. One took 4 hours against roughly 3 months manually; the other contained a single minor error caught by standard human review.
• The new company: The Column Group is founding a startup called Pleiades that will use Claude to run more and more complex drug programs than a small team normally supports. Its first Claude skill, which judges whether drug targets are worth pursuing, was evaluated on 45 target-disease pairs.
• Bristol Myers Squibb: The company is using Claude to assess early-stage clinical trial plans and estimate probability of success. Michael Pourdehnad, who leads early clinical development for hematology, oncology and cell therapy, said that from a single prompt Claude produced a robust initial analysis that would otherwise have taken days.

Background

CAR T therapy has historically been used against blood cancers, and extending it to solid tumors is a long-standing industry problem. The Generate work targets that gap by shortening part of the therapy's genetic instructions.

Between the lines

The pattern across the six partnerships is less about a single breakthrough than about headcount. A one-person company coordinating 11 labs, six scientists generating viable CAR T designs, and a 25-hour review compressed to under 3 all describe the same shift in the team size a drug program requires. Anthropic frames the limits honestly: most of this work remains preclinical, and the Enveda draft still needed human correction.

What's next

Anthropic and Bristol Myers Squibb plan to expand their work in 2027 to early development success probabilities. Anthropic says it will extend Claude into clinical trials. The Mithril candidate could reach human trials by late 2028 at the earliest.

Source: Anthropic