Anthropic Details Claude's Role in Drug Discovery, Targeting 2028 Clinical Trials

Baoyu (@dotey) · 2026-10-27 20

Anthropic reveals progress on Claude's involvement in new drug development: From design and experimentation to drafting regulatory submissions, with the earliest human clinical trials projected for late 2028.

In a post published on October 27, Anthropic detailed what Claude has accomplished in drug discovery over the past few months: deep collaborations with six pharmaceutical and biotechnology companies, as well as the acquisition of a one-person drug company. To set expectations: these projects are largely still in the laboratory and animal testing phases; no drugs have yet been tested in humans. According to Anthropic, the most advanced project may not begin human clinical trials until late 2028 at the earliest.

Developing a new drug from concept to market typically takes more than a decade, involving target identification, molecular design, iterative optimization, manufacturing scale-up, regulatory submission preparation, and clinical trials. The commonly cited "AI-designed molecules" represent only one step in this process. Anthropic aims for Claude to participate in every stage.

[What Claude Did Across Various Companies]

For CAR-T cancer therapies—which involve equipping patients' immune cells with a genetic "instruction manual" to recognize cancer cells—scientists at Generate Biomedicines used Claude Code to shorten a segment of this manual by approximately 70%. This freed up space to add other instructions, potentially enabling treatment of solid tumors in the future. In the lab, the new version performed even better, increasing the number of "probes" (receptors) on the surface of immune cells that recognize cancer cells by about 10 times.

Divergence Labs is developing a drug for food allergies, requiring a molecule capable of dislodging IgE antibodies (which trigger allergic reactions) from immune cells. Claude handled this task almost autonomously, generating approximately 35,000 designs in one month. Among the screened candidates, three showed stronger binding affinity than existing drugs.

BigHat Biosciences asked Claude to propose modifications for two candidate drugs. Over 10 weeks, they created and tested approximately 1,010 variants. The best versions improved target binding strength by more than 10 times, while one variant reduced aggregation speed by over 100 times, making it easier to manufacture and store.

At Enveda, production chemists spent up to 75% of their time on paperwork. Claude reviewed 1,061 pages of production documents, identifying all 11 issues found by humans plus seven additional errors missed by staff. Review time dropped from over 25 hours to under 3 hours. For a critical regulatory document, Claude produced a draft in 4 hours—a task that would take humans approximately 3 months.

Additionally, Bristol Myers Squibb (BMS) is using Claude to evaluate clinical trial protocols for early-stage drugs. Investment firm The Column Group is establishing a new company powered by Claude, designed to enable small teams to advance multiple drug programs simultaneously.

[The One-Person Drug Company]

Cody Tranbarger founded Mithril Medicine, a solo-run drug company researching rare pediatric muscular dystrophy. He delegated contracts, logistics, and coordination with 11 external laboratories to Claude. Within less than 18 months, he moved a therapeutic concept for one disease from idea to efficacy in mice. Anthropic acquired the company (undisclosed amount), and Cody joined Anthropic to continue his work.

The condition is LAMA2 congenital muscular dystrophy. Affected children are born lacking a protein that anchors muscle fibers, causing muscles to degenerate from birth. According to research cited in the technical report, only 1 out of 42 children completely deficient in this protein can walk independently, and there are currently no therapies that alter the disease course. Mithril designed a bispecific antibody to re-anchor a related residual protein in the muscle back onto the muscle structure. In mouse experiments, the treated group showed roughly double the forelimb strength and grip force compared to the untreated group.

[How Reliable Are These Results?]

All results come from partners and Anthropic’s own technical reports and preprints; the article does not mention independent external verification. Mithril’s report notes that it has not undergone peer review. The mouse study was a pilot experiment with 7 treated subjects and 4 controls. When comparing by litter, the differences did not quite reach standard statistical significance thresholds. The report itself acknowledges that rigorous replication studies are needed to confirm the findings.

The report also details Claude’s mistakes: authors identified six errors in Claude’s week-long workflow. Five were caught by Claude itself, but at least three had already been sent via emails after human review. Two of these errors went unnoticed for over four days. Claude only drafts emails; sending decisions remain with humans.

Original Post: https://t.co/P5Bw0mYzJ5 Mithril Technical Report: https://t.co/CtBsbz4vAi

↳ Quoting @AnthropicAI (Anthropic): Over the past few months, Anthropic scientists have engaged in deep research partnerships with six organizations developing new medicines. We've also begun drug discovery efforts of our own.

In doing so, we’ve learned much more about how Claude can support the many stages of drug development.

In this blog post, we share some early results from these efforts, and describe how this work fits in with our wider ambitions in the life sciences.