Early progress on accelerating drug development with Claude

What it claims

Anthropic reports early results from embedding Claude in real drug programs at six partners and in its own acquired company, Mithril Medicine. The work spans five stages: modality engineering, hit generation, lead optimization, manufacturing paperwork and company creation. Anthropic's thesis is that drug-development bottlenecks are spread across the whole pipeline rather than concentrated in molecule design. It says it has already prioritized specific model improvements based on these programs, but it does not name them. Each case links to a partner technical report or preprint, and those reports hold the methods.

Evidence

Generate Biomedicines (CAR T for solid tumors). Six scientists used Claude Code to redesign part of the CAR T genetic construct. One design was about 70% shorter than the standard version. In lab tests, cells carrying it made roughly 10x as many receptors, killed tumor cells more efficiently and stayed active longer.

Divergence Labs (food allergy, bispecific antibody). Claude turned the company's goals into a design plan. It then generated about 35,000 antibody designs in about a month, largely autonomously. Early tests found 3 designs that bind IgE and 20 that dislodge it from immune cells. Three of the 20 appeared to dislodge more IgE than the standard medication. The two arms have not yet been combined into one drug.

BigHat Biosciences (lead optimization). Claude was connected to BigHat's lab data system, Reccy. It proposed edits to two antibody fragments, and the lab made and tested about 1,010 variants in about 10 weeks. Partway through, Claude and an Anthropic researcher judged the standard heat-stability and aggregation metrics misleading. They reanalyzed the raw data and proposed new metrics plus a protocol change, and the main predictions held when BigHat ran it. Against the original leads, the best variants bound more than 10x more tightly. The cancer-targeting lead stayed in solution at a temperature 18°C higher. The half-life-extension lead clumped more than 100x more slowly.

Enveda (manufacturing records and regulatory filings). Manufacturing chemists had spent up to 75% of their time on paperwork. Enveda connected its manufacturing documents to Claude and had it flag issues across 1,061 pages. Claude caught all 11 issues that human reviewers found and flagged 7 more that they had missed. Review time fell 89%, from over 25 hours to under three. Claude also drafted two regulatory documents. One took four hours, against about three months by hand. Enveda's standard review found one minor error in the other.

The Column Group / Pleiades. TCG is founding Pleiades to run more, and more complex, programs per small team using Claude. Its first Claude skill assesses whether drug targets are worth pursuing. It has been evaluated on 45 target-disease pairs, and the page gives no accuracy result.

Bristol Myers Squibb. BMS uses Claude to assess clinical-trial plans for an early-stage drug and to estimate the trials' chance of success. BMS's Michael Pourdehnad says a single prompt produced an initial analysis that would otherwise take days. No quantitative result is given. Both companies plan to expand the work in 2027.

Mithril Medicine. Founder Cody Tranbarger used Claude to coordinate contracts, logistics and experiments with 11 external labs. The company took a muscular dystrophy candidate from idea to promising mouse results in under 18 months. Anthropic expects human trials could start as soon as late 2028 if the candidates hold up.

How to use it

The page names four access routes. Claude Science is an AI workbench for researchers. Anthropic also offers Claude credits for life scientists. The Life Sciences Verification Program gives vetted organizations its most capable models with more permissive biology safeguards. An open call invites guest investigators to propose drug-discovery research. The integration patterns are Claude Code for design work, a connection to a lab data system (Reccy) and a document-corpus connection for quality review. Details are in the partner technical reports, not on this page.

Limits

Anthropic states the work is early and mostly preclinical. No drug from these programs has been tested in humans. The Divergence and Generate results are early lab tests. Pleiades and BMS report no measured outcomes. Several figures are approximate ("about", "roughly"). The page does not say which Claude model or version was used. The page is an Anthropic update about its own product and partners, and independent replication is not reported.