Anthropic puts Claude to work across drug development, with early lab results

Anthropic says it has partnered with six organisations developing medicines and begun drug-discovery work of its own, using Claude across tasks from molecule design to manufacturing paperwork. Early partner results include lab-tested candidates and a reported 89% reduction in review time, but the work is mostly preclinical, not evidence of approved treatments or benefits for patients.

Anthropic Watch analysis

What happened

In a post published on 27 October, Anthropic described work with BigHat Biosciences, Bristol Myers Squibb, Divergence Labs, Enveda, Generate Biomedicines and The Column Group. It also says it has acquired Mithril Medicine, a one-person company developing treatments for rare forms of muscular dystrophy.

The examples range from Claude generating antibody designs for Divergence to helping Enveda review manufacturing documents. Anthropic says it is also working with Bristol Myers Squibb on clinical-trial planning, with plans to expand that work in 2027. Its own drug-discovery programmes include rare childhood diseases.

Key findings

• Generate tested a shorter CAR T design
  Generate says one design was about 70% shorter and, in lab tests, enabled immune cells to make roughly 10 times as many receptors.
• Divergence generated about 35,000 antibody designs
  The company found early “hits” for both parts of its food-allergy drug concept; three designs appeared to dislodge IgE better than the standard medication.
• BigHat tested about 1,010 variants
  Anthropic says the strongest versions of two antibody leads bound more than 10 times as tightly, with other measures of stability also improving.
• Enveda reported faster document review
  Across 1,061 pages, Claude caught all 11 issues human reviewers had identified and flagged seven more; review time fell 89%, from over 25 hours to under three.
• Mithril’s research remains preclinical
  Anthropic says a candidate reached promising results in mice in under 18 months. It expects a candidate could be ready for human trials as soon as late 2028 if results hold up.

Why it matters

This is a more grounded pitch for AI in medicine than a model producing a plausible-looking molecule and calling it a day. The examples describe work tied to real drug programmes, laboratory testing and manufacturing records, across several stages where development can stall.

But the results are early and come from Anthropic’s account of its partnerships. Most of the work is preclinical, and promising laboratory findings are not proof that a medicine will work safely in people. That distinction is rather more important than the word “breakthrough”.

Our read

The most interesting point is the breadth: Claude is being used not just to propose designs, but to support testing, paperwork and clinical-trial planning. If those results stand up to scrutiny, AI may help teams spend less time on bottlenecks and more time deciding which candidates deserve further work.

For now, treat this as a substantial progress report, not a cure announcement. The useful next evidence will be technical detail, replication and results as programmes advance towards human trials.

What to watch

• Whether the partner technical reports provide enough detail for outside researchers to assess the methods.
• How the reported results hold up in further laboratory testing.
• Whether the Bristol Myers Squibb work expands as planned in 2027.
• If any candidate reaches human trials, and what those trials show.

Discussion spark: Should AI drug-discovery claims be judged mainly by early laboratory results, or should companies wait for evidence from human trials before describing the work as progress?

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.