Anthropic shares drug-development case studies with Generate Biomedicines

Anthropic published six drug-development case studies on October 27, 2026. One involves Generate Biomedicines, which used Claude Code to propose new designs for part of a CAR T-cell therapy. In lab tests, the redesigned version was more effective than the standard one, allowing immune cells to make roughly 10 times as many receptors, the proteins they use to spot cancer cells. Turing Post reported the same figure in a brief note in its November 2 newsletter.

CAR T-cell therapy works by giving a patient's immune cells genetic instructions for receptors that recognize cancer. The therapy has historically treated blood cancers, and Generate aims to adapt it for solid tumors. Anthropic's post says the approach shortens one part of those instructions, which frees room for additional instructions suited to other kinds of cancer. Six Generate scientists used Claude Code to propose designs for that part. One was about 70% shorter than the standard version. Beyond the receptor count, Anthropic says the modified cells killed tumor cells more efficiently and stayed active for longer.

Generate's team posted a bioRxiv preprint on October 29, 2026, titled Refactored intracellular domains enhance CAR expression and function. The intracellular domain is the part of the receptor that couples antigen recognition to T-cell activation. The preprint describes an AI-agent process run over three rounds, in which the agents proposed designs and refined them as measurements came in. The result, krgr-ζ1-40, is a 45-amino-acid domain, less than a third the length of the clinical 4-1BB-CD3ζ (BBζ) domain. Compared with BBζ, the abstract says it increased CAR surface expression and antigen-driven IFN-γ release, without detectable antigen-independent IFN-γ release. It also reports extended expression and target-cell killing after transient mRNA delivery, and enhanced B-cell depletion with lentiviral and lipid-nanoparticle delivery. All nine authors are listed with Generate affiliations, and the competing interest statement says none were declared.

The two documents describe the work at different levels. The abstract uses the phrase AI agents and does not name Claude. It reports the expression gain without a fold-change figure, so the roughly 10-fold number appears only in Anthropic's post. Anthropic itself says the broader partnership work is mostly preclinical, meaning the drugs have not yet been tested in humans. The post also does not say how many design rounds the Claude-proposed versions went through in total.

The other case studies cover different stages of drug development. Divergence Labs had Claude turn its goals into a design plan. Claude then generated roughly 35,000 antibody designs in about a month, and the company found three designs that could bind IgE and 20 that could dislodge it. At Enveda, Claude checked 1,061 pages of manufacturing documentation, caught all 11 issues human reviewers had found and flagged seven more, and cut human review time by 89%, from over 25 hours to less than three. BigHat Biosciences, Bristol Myers Squibb, and The Column Group, which is setting up a company called Pleiades, round out the partners. Anthropic also reported its own program for rare childhood muscular dystrophy, which it says could be ready for human trials as soon as late 2028 if the candidates continue to hold up.