Claude for Drug Discovery: Anthropic's Early Results Explained

Anthropic says Claude has helped six external drug-development partners speed up parts of the slow, expensive process of turning a biological idea into a tested medicine, and that the company has now started its own in-house drug discovery programs. The update, published on Anthropic's research blog on October 27, 2026, covers work across modality engineering, antibody design, lead optimization, manufacturing paperwork, company formation and clinical trial planning — and includes the acquisition of a one-person biotech startup now working inside Anthropic.

This is not a product launch. It is a progress report on where Claude helped scientists move faster, where its outputs still needed heavy human correction, and what Anthropic plans to do next. Every figure below comes from Anthropic's own post; none of it has been independently tested by Pandromeda, and none of the programs described have reached human clinical trials.

Why drug development is a target for AI labs

Getting a single new medicine from an initial idea to a pharmacy shelf is one of the slowest and most expensive processes in any industry. Published cost estimates vary widely depending on methodology, but independent analyses have put the fully loaded cost of bringing one approved drug to market anywhere from under $1 billion to well over $2.5 billion once the cost of failed candidates is included, and the overall process commonly spans a decade or more, according to the Wikipedia summary of cost-of-drug-development research. That combination of years-long timelines, heavy regulatory documentation and high failure rates is exactly the kind of bottleneck Anthropic says it is trying to attack with Claude — not by replacing lab science, but by compressing the analysis, paperwork and decision-making that surrounds it.

What Anthropic announced

Anthropic's life sciences team spent several months running what it calls deep research partnerships with six organizations developing new medicines: Generate Biomedicines, Divergence Labs, BigHat Biosciences, Enveda, The Column Group and Bristol Myers Squibb (BMS). Anthropic's framing is that drug development is bottlenecked at nearly every stage — not just the much-hyped step of designing a molecule — so it tested Claude across a wider range of tasks: engineering cell therapies, generating antibody candidates, reanalyzing lab data, drafting regulatory filings, evaluating drug targets, and assessing clinical trial plans.

Alongside the partner work, Anthropic disclosed that it has acquired Mithril Medicine, a company founded and, until the acquisition, run by a single person, Cody Tranbarger, to develop treatments for two rare forms of muscular dystrophy. Tranbarger now leads that research inside Anthropic.

Generate Biomedicines: shortening a CAR T-cell therapy

Generate Biomedicines is working to extend CAR T-cell therapy — currently used mainly for blood cancers — to solid tumors, in part by shortening the genetic payload that has to be packaged into cells. Six of the company's scientists used Claude Code to propose redesigns. Anthropic reports that one resulting construct was roughly 70% shorter than the standard version while performing better in lab testing: about 10 times more receptor expression, more efficient killing of tumor cells, and longer-lasting immune-cell activity. Generate Biomedicines has published its own account of the work.

Divergence Labs: 35,000 antibody designs for a food-allergy treatment

Divergence Labs is developing a food-allergy treatment built around a two-arm antibody designed to detach IgE antibodies from immune cells, which is the mechanism behind allergic reactions. According to Anthropic, Claude translated the program's scientific goals into a design plan and then generated about 35,000 candidate antibody designs over roughly a month. Of those, three were confirmed binders and 20 designs could dislodge IgE at all; three of those 20 reportedly outperformed the standard allergy medication at that specific task. The next step for the program is combining the two antibody arms into a single working molecule. Divergence Labs has posted a preprint describing the IgE work.

BigHat Biosciences: reanalyzing lab data to pick better antibody leads

BigHat connected Claude directly to its laboratory data system and had it propose changes to two antibody-fragment drug leads. Roughly 1,010 variants were synthesized and tested over about ten weeks. Anthropic says Claude, working with an Anthropic researcher, also reanalyzed the resulting data and suggested new stability and aggregation metrics, which BigHat validated against its own methods. The reported gains over the original leads: more than 10 times tighter target binding, an 18°C increase in solution stability for the cancer-targeting lead, and more than 100 times slower clumping (aggregation) for the longer-acting lead. BigHat published its own collaboration report alongside Anthropic's post.

Enveda: cutting regulatory review time by 89%

Enveda's case is less about molecule design and more about the unglamorous paperwork that slows manufacturing and regulatory teams down. Anthropic says the company's manufacturing chemists had been spending up to 75% of their time on documentation rather than lab work. In testing, Claude reviewed 1,061 pages of manufacturing documentation, caught all 11 issues that human reviewers had separately found, and flagged seven additional issues. Human review time on that material reportedly dropped by 89%, from more than 25 hours to under three. Claude also drafted two regulatory documents from scratch: one took about four hours versus roughly three months by hand, and the other needed one minor correction during standard human review. Enveda describes the underlying tool, its CMC Document Processor, on its own site.

The Column Group: building a company run on Claude skills

The Column Group (TCG), a biotech investment firm, is using the partnership to help stand up a new company, Pleiades, intended to run an increasing number of complex drug programs with a small human team supported heavily by Claude. Its first purpose-built Claude skill assesses whether a given drug target is worth pursuing for a given disease, and Anthropic says it has been tested on 45 target-disease pairs so far. This is the most structurally ambitious piece of the announcement: rather than using Claude as a tool inside an existing drug company, TCG is trying to design an organization around it from the start.

Bristol Myers Squibb: judging clinical trial plans

BMS used Claude to assess early clinical trial plans for an early-stage drug and estimate its probability of success. Michael Pourdehnad, MD, BMS's Senior Vice President and Head of Early Clinical Development, is quoted in Anthropic's post describing Claude producing "a robust initial analysis" from a single prompt — work that would otherwise take a team days. Anthropic says BMS plans to expand this use in 2027, including extending it to estimating probability of success earlier in development.

Anthropic's own drug discovery: the Mithril Medicine acquisition

Separately from the six partnerships, Anthropic disclosed it has acquired Mithril Medicine, a company founded by Cody Tranbarger to develop treatments for two rare, childhood-onset forms of muscular dystrophy. Before the acquisition, Mithril was essentially a one-person operation: Tranbarger used Claude to coordinate contracts, logistics and experiments across 11 external labs, delegating much of the day-to-day operational work to the model. Anthropic says the team moved a candidate treatment from early idea to promising results in mice in under 18 months, and that Tranbarger now leads this research inside Anthropic with plans to start new programs. If the results hold up through further testing, Anthropic says a candidate could be ready to enter human clinical trials as early as late 2028 — a timeline the company is explicit is not guaranteed.

What this does and doesn't show

Every program described is preclinical. Anthropic is direct about this: none of the molecules, antibodies or treatment candidates covered in the post have been tested in humans, and standard drug development includes years of further toxicology, manufacturing scale-up and clinical trials before anything resembling a product exists. The results are also self-reported by Anthropic and its partners rather than independently verified or peer-reviewed in the post itself. Several of the headline figures — the 35,000 antibody designs, the 89% paperwork reduction, the 10x binding improvement — are process efficiency metrics (how fast a task got done, or how much a lab-measured property improved in testing) rather than evidence that any resulting drug will work in patients. For context on how Anthropic positions this work relative to its wider research push into scientific applications, it previously expanded its AI-for-Science program to 20 research partners, and the drug-development partnerships described here sit alongside that broader effort rather than replacing it.

How Anthropic wants other labs to use this

Anthropic says it plans to scale this work "substantially" and is opening up access through several channels: Claude Science, which it describes as an AI workbench for scientific work; Claude credits earmarked for life scientists; a Life Sciences Verification Program that gives vetted, trusted organizations more permissive safeguards than Claude's default consumer settings; and an open call for guest investigators to work with the life sciences team directly. Anthropic is also inviting other drug discovery organizations to apply through a contact form on its site. The tools involved sit within Claude's broader life-sciences product push — the kind of specialized workbench approach Anthropic has also taken with developer-facing tools like Claude Code, repurposed here for lab environments rather than software repositories.

What's next

Anthropic names two concrete next steps. First, it wants to extend Claude's role further into clinical trials as the existing programs mature, and to bring on partners that are already at the clinical stage rather than purely preclinical. Second, it is widening access to the tools behind this work — Claude Science, dedicated life-sciences credits, the Verification Program and the guest-investigator call — so that organizations beyond the initial six partners can test similar workflows. BMS's clinical-trial-assessment work is explicitly slated to expand in 2027, and Anthropic's own Mithril Medicine program is targeting a possible human trial entry as early as late 2028, contingent on further results. None of these are commitments to specific outcomes; they are stated intentions in a company blog post about ongoing, unfinished research.

Written byTheo Park

Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

More from Theo ParkAll AI coverage