Anthropic shares early results from Claude in drug development: six partners including BMS, plus its own programs via Mithril Medicine

Background and industry pain points

Drug development is slow, expensive and failure-prone, so investors tend to favor validated ideas. Talk about AI in pharma usually focuses on proposing first molecule designs, but the real bottlenecks are spread across target selection, design, manufacturing scale-up, regulatory filings and trial design. Anthropic argues there is no single problem to solve, so its life sciences team spent months embedded in real programs with six partners and started drug discovery work of its own.

How the work is structured

Generate Biomedicines used Claude Code to redesign part of the genetic instructions for CAR T-cell therapy, aiming to extend it to solid tumors. At Divergence Labs, Claude turned goals into a design plan and generated antibody candidates largely on its own. BigHat connected Claude to its lab data system, Reccy, for lead optimization, where Claude and a researcher found the standard stability and clumping metrics misleading and proposed new ones. Enveda connected its manufacturing documents to Claude to flag issues and draft regulatory documents. The Column Group is founding Pleiades, a company that will use Claude to run more drug programs with a small team, and BMS is using Claude to assess early clinical trial plans and estimate their odds of success, with plans to expand in 2027.

Reported results

All figures are vendor-reported and not independently reproduced: about 35,000 antibody designs in about a month at Divergence; about 1,010 variants tested in about 10 weeks at BigHat with over 10x tighter binding; at Enveda all 11 known issues caught plus 7 more across 1,061 pages, 89% less review time, and one regulatory draft in 4 hours versus about 3 months by hand; and a CAR-T component about 70% shorter with roughly 10x receptor expression at Generate. These are project-level lab and workflow metrics rather than general AI benchmarks, and most of the work is still preclinical.

What developers can take from it

For AI-for-science builders, the useful pattern is wiring Claude into lab data systems and document stores and letting it help define metrics, not just propose candidates. Anthropic also points to Claude Science, its research workbench, Claude credits for life scientists, the Life Sciences Verification Program with more permissive safeguards for trusted organizations, and an open call for guest investigators. Partner technical reports and preprints are linked from the official post.