Anthropic's Six Pharma Partnerships Produce First Lab Results: 35,000 Antibody Designs, 10x CAR T-Cell Receptors
Anthropic published its first progress report from a set of deep research partnerships with six drug development organizations. The company’s life sciences team has been embedded in active drug programs for several months, running Claude as a working scientific agent across discovery, design, and regulatory preparation tasks. Early lab results are now in for multiple programs across the cohort.
Partners: BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines, and The Column Group. Anthropic is also running its own internal drug discovery programs, including a rare childhood muscular dystrophy program where Claude handles most of the scientific and operational work.
The Two Programs With Lab Data
Generate Biomedicines — CAR T-cell therapy for solid tumors
CAR T-cell therapy has historically been effective for blood cancers. Generate Biomedicines is working to extend it to solid tumor cancers, which requires shortening one segment of the therapy’s genetic instructions to create space for additional tumor-targeting code.
Six Generate scientists used Claude Code to propose new designs for this segment. One design came in approximately 70% shorter than the standard version. In lab tests, that design allowed immune cells to produce roughly 10 times as many receptors — the proteins cells use to detect cancer — compared to the standard. The redesigned cells killed tumor cells more efficiently and remained active for longer.
Divergence Labs — food allergy drug
Standard food allergy medications neutralize IgE circulating in the bloodstream. Divergence Labs is developing a drug that would detach IgE from immune cells entirely, a mechanistically different approach that could provide more durable relief.
The design challenge is precise: the molecule needs two distinct arms, one that binds tightly to IgE and one that physically prises IgE off the immune cell. Claude converted Divergence Labs’ requirements into a detailed design plan and then generated approximately 35,000 antibody candidates in roughly one month, largely autonomously.
Early testing found hits on both arms: three designs that bound to IgE and 20 that could dislodge it. Three of the 20 dislodging candidates performed better than the standard reference medication in early assays.
What Anthropic Is Actually Claiming
The company is careful on scope. These are early-stage results — in vitro lab data, not clinical outcomes. Anthropic frames the partnerships as a way to identify where models fall short in actual drug development workflows and to prioritize specific model improvements based on what gets found. It also treats the programs as case studies for what would be required to bring Claude into active use at more biotech firms.
The three goals Anthropic names for the program: learn where current models fail at real drug tasks, demonstrate what AI involvement in a live program looks like for the broader industry, and produce direct impact in the targeted disease areas if the results continue positively.
Disease areas across the six programs include food allergies, rare childhood diseases, and several forms of cancer.
Context
Anthropic published technical reports and preprints alongside the blog post describing methods and early results for each partner. The company says it plans to scale this work substantially and has already reprioritized specific model improvements based on what it learned through the partnerships.
The Generate Biomedicines result is notable because CAR T-cell therapy extension to solid tumors is a well-recognized bottleneck in oncology. Getting a 10x receptor improvement from a redesigned genetic component in lab conditions is a concrete early result, even if the distance to a clinical application remains long.