A November 6 report said Generate Biomedicines was taking its Claude-assisted CAR-T signaling-domain design into animal models. The work began with laboratory experiments on a 45-amino-acid candidate, krgr–ζ1–40, designed with Claude agents and Generate’s protein-generation models. Fierce Biotech reported the animal-model development.

The reported results are laboratory signals from a design effort, not a demonstrated benefit for patients.

How Claude entered the design process

The target was the CAR-T cell’s intracellular domain (ICD), the signaling component that links antigen recognition to T-cell activation. Generate paired its protein-generation models with Claude agents, while its biologists reviewed the proposed sequences and fed experimental measurements back into the work. Anthropic said six Generate scientists used Claude Code to propose designs. The preprint describes three rounds of design and measurement feedback.

What the candidate showed in laboratory experiments

The preprint names the design krgr–ζ1–40 and gives its length as 45 amino acids, less than one-third the length of the clinical 4-1BB–CD3ζ (BBζ) intracellular-domain comparator. Its abstract reports increased CAR surface expression and antigen-driven IFN-γ release compared with BBζ, with no detectable antigen-independent IFN-γ release.

The abstract also reports different results across delivery contexts:

  • After transient messenger RNA (mRNA) delivery, the design was associated with extended CAR expression and target-cell killing. It also supported expression of four CARs from a shared mRNA dose.
  • After lentiviral and lipid-nanoparticle delivery, the authors reported enhanced B-cell depletion.

Anthropic’s account gave a separate summary: it described a design as about 70% shorter than a standard version and said laboratory tests produced roughly 10 times as many receptors. Those figures are Anthropic’s characterization; the preprint’s comparison is specifically with clinical BBζ.

The findings remain preclinical

bioRxiv labels the work a preprint that has not been certified by peer review. The reported findings are laboratory results and do not establish clinical benefit for patients.

On November 6, Fierce Biotech reported that Generate was taking the work into animal models. Generate CEO Mike Nally hoped to have data by year-end, according to the report.