Anthropic's drug development results: six partners, strong preclinical numbers, no human data

Anthropic has published its first batch of results from working with drug developers. In a post dated October 27, 2026, the company's life sciences team described work with six organizations: BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group. It also said it has started its own drug discovery programs. If you are trying to judge whether Claude drug development results are real, the short answer is that they are specific and unproven in patients.

All figures below come from Anthropic's post unless stated. None has been independently evaluated, and the post says most of the work is preclinical, meaning not yet tested in humans.

The lab results

Three case studies involve wet-lab measurements, which are the most credible kind in the post.

At Divergence Labs, working on food allergies, Claude turned the team's goals into a design plan and generated roughly 35,000 antibody designs in about a month. The team found three designs that bind IgE and 20 that dislodge it from immune cells. Three of those 20 dislodged IgE more effectively than standard medication. The next step is combining the two functional arms into a single drug.

At BigHat Biosciences, Claude connected to the company's lab data system and helped improve two antibody fragment leads. About 1,010 variants were made and tested over roughly 10 weeks. According to Anthropic, Claude and an Anthropic researcher reanalyzed the raw data, found the standard heat-stability and clumping metrics misleading, and proposed new ones. The best versions bound more than 10 times more tightly, one lead stayed in solution 18°C higher, and another clumped more than 100 times more slowly.

At Generate Biomedicines, six scientists used Claude Code to propose redesigns of a genetic component for CAR T-cell therapy. One design was about 70% shorter than standard. In lab tests, immune cells made roughly 10 times as many receptors, killed tumor cells more efficiently and stayed active longer.

The paperwork results

The least glamorous result may be the most reproducible. At Enveda, manufacturing chemists spent up to 75% of their time on paperwork. On 1,061 pages of documentation, Claude caught all 11 issues that human reviewers found and flagged seven more they had missed. Human review time fell 89%, from over 25 hours to under three. Claude also drafted two regulatory documents; one took four hours versus about three months by hand, and human review of the other found one minor error.

That is a credible use. Document review has a known answer key, errors are countable, and a human still signs. It also does not touch the hard part of medicine, which is whether a molecule works in people.

Planning, targets and a one-person company

Bristol Myers Squibb uses Claude to assess early-stage clinical trial plans and estimate their likelihood of success, and the partners plan to expand that in 2027. BMS's Michael Pourdehnad, MD, is quoted saying Claude "generated a robust initial analysis that helped inform our assessment". Note the verb: informed, not decided.

The Column Group is creating a company called Pleiades that will use Claude to run more and more complex drug programs than a small team normally could. Its first Claude skill, which assesses whether drug targets are worth pursuing, was evaluated on 45 target-disease pairs. The post does not give the score.

The post also describes Mithril Medicine, which Anthropic acquired. Its founder used Claude to organize contracts, logistics and experiments with 11 external labs. A possible treatment for a rare muscular dystrophy went from idea to promising mouse results in under 18 months, and, if it keeps holding up, could be ready for human trials as soon as late 2028.

What the post leaves out

Several things are missing, and a skeptical reader should name them. The post links no raw data or methods section for any partner. It reports no comparison to a team using a rival model, or to a team using no model. The BigHat and Divergence figures are best-of-batch outcomes from tens of thousands of designs, not hit rates. And the Mithril timeline is a founder's account of a project the buyer now owns.

Anthropic frames the next step as clinical work and broader access: a Claude Science workbench, Claude credits for life scientists, a Life Sciences Verification Program offering more permissive safeguards to trusted organizations, and an open call for guest investigators. The call, published October 22, offers 8 to 12 weeks of collaboration at no charge and says results will be published with data and code open-sourced.

Why this belongs in the race coverage

Biology is where capability and safety meet. The same permissive access that helps a verified pharma team is the kind of access that safeguards are built to restrict. That is why the verification program matters, and why independent evaluators are central to Anthropic's safety claims, as we reported in the METR access story. It also fits a company moving toward the public markets with a story about real-world impact, which we covered in our report on the public S-1.

Our take

The post is better than most vendor science: numbers, named partners, a limits section. The right posture is interested skepticism. We would want peer-reviewed data from at least two partners, an account of failures, and the first clinical readout before calling this a change in how medicines get made. The Enveda review is the result we would test first, because anyone with a document backlog can check it.