AI and security briefsModels

What did Anthropic show about Claude in drug development, and what does it not show?

By Brandon Da Costa, founder of Digiton Dynamics

Published 28 October 2026Updated 28 October 2026Editorial policy

Anthropic says Claude proposed antibody and cell therapy designs, checked manufacturing records and drafted regulatory documents with six partners, and it reports lab and review-time results. Almost all of it is preclinical and no drug is in human trials. For software teams, the useful part is how the long workflows were set up.

Who should act: Teams that run Claude on long, multi-step work with real data, such as research, compliance or document review, and want a measured example of what worked.

What changed

On 27 October 2026 Anthropic published an update on its life sciences work (Anthropic research post, 27 October). Over several months its scientists worked with six organizations: BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group. Anthropic has also started drug discovery programs of its own. The results below are Anthropic's reporting of what its partners found.

• Generate Biomedicines. Six scientists used Claude Code to propose designs for one part of a CAR T-cell therapy. One design was about 70% shorter than the standard version, and in lab tests it let immune cells make roughly 10 times as many receptors (Anthropic research post, 27 October).
• Divergence Labs. Claude generated roughly 35,000 antibody designs in about a month, largely autonomously. Early tests found three designs that bound IgE and 20 that dislodged it.
• BigHat Biosciences. Claude, connected to the lab data system, proposed changes to two leads. About 1,010 variants were made and tested in about 10 weeks. The best versions bound more than 10x more tightly, one stayed in solution 18°C higher and one clumped more than 100x more slowly.
• Enveda. On 1,061 pages of manufacturing records Claude caught all 11 issues human reviewers had found and flagged seven they had missed. Review time fell 89%, from over 25 hours to under three.

Anthropic is also working with Bristol Myers Squibb on clinical trial planning, with an expansion planned for 2027, and has acquired Mithril Medicine, a one-person company whose founder used Claude to run work with 11 external labs.

Who it affects

• Teams that build agents for long research or document work. The Enveda and BigHat setups are patterns you can copy, with the limits below (Anthropic research post, 27 October).
• Life sciences organizations. Anthropic says it rolled out Claude Science, offers Claude credits to life scientists and runs a Life Sciences Verification Program that gives trusted organizations more permissive safeguards.
• Anyone who plans to quote these figures. Anthropic calls the work early and mostly preclinical. Its Mithril report, dated 26 October 2026, is marked not peer reviewed (Anthropic technical report, 26 October).

What to do now

1. Read the primary reports before you repeat a number. The post links partner reports, and the Divergence preprint gives a denominator the post leaves out. Of 91 experimentally characterized disruptor designs, 20 were active, and three removed more IgE than omalizumab (Divergence Labs preprint, October 2026).
2. Pick a task with known answers. Enveda scored Claude against issues its human reviewers had already found, then counted the extra ones.

   task: review 1,061 pages of batch records
   ground truth: issues humans already found
   measure: caught, extra flagged, review hours before and after

3. Connect Claude to the system of record. BigHat linked it to its lab data system and Enveda to its manufacturing documents, so nobody pasted files.
4. End every loop in a real test. In the Divergence and BigHat work, a lab result decided which designs mattered.
5. Review the metric as well as the answer. BigHat suspected its heat-stability and clumping metrics were misleading, so Claude and a researcher reanalyzed the raw data. A changed protocol was run and the main predictions held.
6. Record the hours. Enveda's 89% is a before and after review time.

What to watch

• Human data. Anthropic says a Mithril drug could be ready to start human trials as soon as late 2028 if its candidates continue to hold up. Nothing in the post shows a patient result.
• Independent replication. The partner numbers are the partners' own, linked from Anthropic's post.
• The Column Group's Pleiades. Its first Claude skill was evaluated on 45 target-disease pairs, and the post gives no result.
• Bristol Myers Squibb. The planned 2027 expansion has no measured result yet, only a quote from a BMS executive.

Questions people ask

Digiton's production review covers how a long-running agent is scoped and logged, and the checks are on the Claude Code governance page.

How this brief was made: researched from the sources below, drafted with AI assistance and checked against each source on the dates shown. See the editorial policy.

Sources

1. Anthropic, early progress on accelerating drug development with Claude, 27 October 2026, read 28 October 2026. https://www.anthropic.com/research/early-progress-drug-development
2. Divergence Labs, preprint on an IgE disruptor antibody with Claude-enabled design, read 28 October 2026. https://divergencelabs.com/publications/ige-preprint.pdf
3. Anthropic, Mithril Medicine technical report on a bispecific bridging antibody, 26 October 2026, read 28 October 2026. https://www-cdn.anthropic.com/52e1f5d50c47ca3818f56704c27cf19e59b62e77.pdf

More briefs

How these briefs are sourced and corrected. All AI and security briefs