Claude for Life Sciences

Claude for Life Sciences is an industry-specific offering from Anthropic that adapts its Claude models for scientific research and drug development, packaging connectors to laboratory and literature platforms, reusable instruction sets called Agent Skills, and support for biotech, pharmaceutical, and academic users. Announced on October 20, 2025, it followed Claude for Financial Services, which Anthropic launched on July 15, 2025.^[1][2][20] Through 2026 Anthropic extended the effort well past the original connector product, adding healthcare tools, running its own protein design and model-optimization campaigns, opening a trusted-access program for vetted biology researchers, and confirming that it operates a wet biology lab. In September 2026 it introduced an in-house life sciences research group and laboratory, whose early results included a previously uncharacterized phage enzyme system, built around an already known reverse transcriptase, that Claude agents flagged in a large sequence database.^{[13][15][16][22]}

Background

Anthropic released Claude for Financial Services on July 15, 2025, a packaged offering aimed at a single vertical, bundling its models with connectors to data providers and with expanded usage for analysts and portfolio managers.^[3][20] Life sciences was positioned as the next target. The company had recruited Eric Kauderer-Abrams as head of biology and life sciences months before the launch, and he framed the ambition in commercial terms. "We want a meaningful percentage of all of the life science work in the world to run on Claude, in the same way that that happens today with coding," he told CNBC.^[2][4]

The pitch rests on the breadth of work a research organization does between an initial idea and a regulatory filing. Anthropic described Claude for Life Sciences as covering literature review, hypothesis generation, data analysis, and the drafting and review of regulatory submissions, rather than a single narrow task.^[1][2] Anthropic is a public benefit corporation, and the announcement tied the effort to that charter, stating that increasing the rate of scientific progress is part of its mission.^[1]

Launch and capabilities

The offering is built on Claude Sonnet 4.5, the model Anthropic released in late September 2025 and described as its most capable at the time of the launch.^[1] Anthropic reported that Sonnet 4.5 scored 0.83 on Protocol QA, a benchmark that tests understanding of laboratory protocols, against a human baseline of 0.79 and the earlier Claude Sonnet 4 score of 0.74.^[1][5] Anthropic's footnote describes the score as a 10-shot multiple-choice result and refers readers to its Sonnet 4.5 system card; Maginative noted that the announcement itself did not detail the test methodology.^[1][5]

Capabilities are grouped around stages of the research process. For discovery work, Claude can summarize and cite biomedical literature and propose testable hypotheses. For bioinformatics, it can process and analyze genomic data, including through Claude Code, and present results as slides, documents, or notebooks. Through the Benchling connector it can draft study protocols, standard operating procedures, and consent documents. For compliance work it can draft and review regulatory submissions and compile supporting data.^[1][5] Anthropic and outside coverage described demonstrations, such as preclinical dosing analysis and regulatory report generation, that compressed work measured in days into minutes.^[2][6]

At launch the product was available through claude.com and the AWS Marketplace, with Google Cloud Marketplace listed as forthcoming.^[5] To help organizations adopt it, the launch announcement named nine implementation partners: Caylent, Deloitte, Accenture, KPMG, PwC, Quantium, Slalom, Tribe AI, and Turing, alongside the cloud partners AWS and Google Cloud.^[1][7]

Connectors and partners

The connectors let Claude read from systems scientists already use rather than from manually exported files. They are built on the Model Context Protocol (MCP), Anthropic's open standard for linking models to external tools and data. The Benchling integration returns answers with one-click traceability back to source experiments and notebooks, and preserves the customer's existing permissions and audit logs.^[1][8] The following platforms were named as connectors at launch.

The Benchling tie-up was announced jointly, with Benchling issuing its own press release on the launch day, October 20, 2025. Ashu Singhal, Benchling's co-founder and president, said "AI in R&D only works through an ecosystem," and Anthropic's Kauderer-Abrams said that with Benchling's data foundation "Claude helps teams move from research questions to results faster." Access to that connector is limited to Benchling customers who have Benchling AI enabled.^[8]

Agent Skills and use cases

Agent Skills are packaged instruction sets, with any scripts and reference files, that guide Claude through a defined procedure so the same workflow can be reused across an organization. At launch Anthropic shipped one named skill for life sciences, single-cell-rna-qc, which performs quality control and filtering on single-cell RNA sequencing data following scverse best practices.^[1][5] Reported use cases spanned the research lifecycle, from compiling literature reviews and generating hypotheses through genomic analysis to drafting regulatory documentation.^[1][2]

Anthropic also lists existing enterprise connectors that life sciences teams can combine with the science-specific ones, including Google Workspace, Microsoft SharePoint, OneDrive, Outlook and Teams, plus the data platforms Databricks and Snowflake.^[5] These sit alongside Claude for Work enterprise deployments, and Anthropic positioned the offering for roles such as researchers, clinical coordinators, and regulatory affairs managers.^[1]

On January 11, 2026, Anthropic expanded the offering and introduced a companion Claude for Healthcare product with HIPAA-ready infrastructure for providers and payers. The expansion added connectors including Medidata, ClinicalTrials.gov, the bioRxiv and medRxiv preprint servers, Open Targets, the ChEMBL compound database, the Owkin pathology platform, and ToolUniverse. It also added Agent Skills for scientific problem selection, conversion of instrument data to the Allotrope format, bioinformatics bundles covering scVI-tools and Nextflow, and clinical trial protocol drafting.^[9][10]

Experimental biology and trusted access (2026)

Through 2026 Anthropic's life sciences work moved beyond the connector-and-skills product in three directions: research campaigns in which Claude itself carried out molecular design and software engineering work, a separate access program that relaxes biology safeguards for vetted organizations, and an in-house research group with its own laboratory.

Protein design campaigns

On August 18, 2026, Anthropic reported a de novo protein design campaign in which Claude Mythos Preview and Claude Opus 4.8, working inside Claude Science with open-source design and folding tools, designed binders against 16 targets. Adaptyv Bio and Twist Bioscience produced and tested the designs. Anthropic said Claude succeeded against 14 of the 15 targets for which usable data came back, with overall hit rates of 26.7% for Mythos Preview and 22.6% for Opus 4.8 when designing against all targets at once, and 35.1% for Mythos Preview working one target at a time, against the 10% to 15% it described as typical for the field. The campaign ran with minimal human involvement beyond approving access requests and ordering the designs.^[18] Adaptyv, which says it received the designs anonymized and did not know which model produced which protein, put the overall hit rate at 26.8% (354 binders from 1,320 designs) and reported that 95% of the designs expressed.^[19]

On September 17, 2026, Anthropic published a follow-up focused on cost rather than capability. Claude, working within Claude Science, optimized more than 30 open-source models used for biomolecular structure prediction, protein design, protein language modeling and genomics in just under four weeks, making them roughly four times faster on average with a minimal loss of precision, and nearly twice as fast with identical outputs. It also produced a low-memory "Big" mode that allows accurate prediction of biomolecular systems larger than 10,000 tokens on a single NVIDIA GPU node, and a set of GPU kernels called FlashPairformer. Anthropic said the work was supervised by two members of its technical staff who were experienced in biomolecular modeling but had no prior experience in inference optimization or kernel engineering, and it open-sourced the optimized code.^[13] The stated motivation was accessibility: the earlier binder campaign had been allowed to spend up to $10,000 per target on the Modal platform, roughly 2,500 NVIDIA H100 GPU hours, which Anthropic described as more resources than most protein designers can reach.^[13]

Protein design competition with Adaptyv Bio

The same post announced a protein design competition co-sponsored with Adaptyv Bio, with five problems released one per week from September 28 to October 31, 2026, covering challenges such as species cross-reactivity, pH sensitivity and peptide-MHC specificity, and difficult targets including GPCRs.^[13][14] Experimental validation is planned by November 30, 2026, with results published on Adaptyv's Proteinbase on December 15, 2026.^[14]

The two organizers describe the money differently. Anthropic's post says that, with Adaptyv and with contributions from Modal and Twist Bioscience, it is "committing up to $1 million in Claude credits and $250,000 in Modal compute credits, as well as wet lab validation for over 5,000 designs."^[13] The Proteinbase competition page instead says Anthropic and Adaptyv are "jointly sponsoring $1 million+ in experimental validation" to test more than 5,000 designs at no cost to participants, with Anthropic providing "an additional $1 million+ in Claude credits" to selected teams and researchers.^[14] The competition has no cash prize, though special non-cash prizes are promised.^[14]

Entry is split into three tracks. Track 1, for research labs and companies with significant protein design experience, offers up to $50,000 in Claude credits for academic teams and up to $25,000 for industry teams, with about 30 teams to be selected and a reserved testing allocation of up to about 15 designs per challenge. Track 2 gives individual researchers and teams of up to three complimentary Claude Max 20x access during the competition. Track 3 is an open, self-supported track in which Claude use is optional. Applications for Anthropic support in Tracks 1 and 2 ran from September 16 to 24, and designs in Tracks 2 and 3 have no guaranteed testing allocation. Designs chosen for testing are selected with a Claude-based workflow, and results for tested designs, including negative results, are to be published on Proteinbase. The page says targets and requirements are published as each problem opens, with the first submission window scheduled for September 28 to October 4, 2026.^[14]

Life Sciences Verification Program

Also on September 17, 2026, Anthropic introduced the Life Sciences Verification Program (LSVP), which gives verified life science professionals access to its Mythos, Opus and Sonnet models under safeguards that are more permissive for biology than those applied to generally available models. Anthropic said it had already onboarded dozens of organizations through an early-access program and was opening applications to the broader community, and it described the program as launching in beta that day, initially for teams and institutions rather than individual Pro and Max plans.^[13][15] It is aimed at work that generally available Fable models block, such as drug discovery, research biology, clinical development and manufacturing.^[15]

Applicants are vetted on research credentials, security standards and ethical research oversight. Verified teams can then apply for two kinds of grant.^[15]

Anthropic said it was working with the US government on broader high-risk access to Mythos models. Safeguards outside life sciences, such as cyber classifiers, stay in place under both grants.^[15] The program also changes how enforcement works: rather than blocking requests in real time, Anthropic monitors LSVP traffic offline against the use cases an organization declared in its application, and it requires 30-day data retention for that traffic. It says the retained data is compartmentalized, cannot be used for model training, and is not accessible to its own life sciences research teams.^[15] At launch LSVP was available in Anthropic's first-party console for API use and in Claude for Enterprise and Team plans, but not on individual plans, not on third-party platforms, and not for organizations with a business associate agreement in place.^[15] The announcement quoted statements from Xaira Therapeutics, Edison Scientific and Manifold Bio.^[15]

When Anthropic released Claude Opus 5.5 on September 22, 2026, it said that because the model is comparable to Claude Mythos 5.1 in biology and cybersecurity, it was deploying it with safeguards similar to those on Claude Fable 5.1, and that vetted organizations could apply to the LSVP to use Opus 5.5 for biology research.^[17]

Wet lab

On September 18, 2026, Reuters reported that Anthropic had set up a wet lab for physical biology work in the San Francisco Bay Area, citing two people familiar with the matter, and TechCrunch reported the same day that Anthropic had confirmed the lab to it.^[21][16] In an interview with Reuters earlier that week, Eric Kauderer-Abrams, Anthropic's head of life sciences, said, "We believe that to do biology, the final test is still and will be for a while in real lab work," adding that "we absolutely are doing that today." He described the approach as typical of most biotech companies, with some work done in Anthropic's own facilities and some with external partners.^[21] A spokesperson said the lab was not for drug discovery specifically, and TechCrunch reported that the company named fundamental biology as the main focus, linking that to Anthropic's reluctance to appear to compete with the pharmaceutical companies that are among its customers and partners.^[21][16] Kauderer-Abrams told Reuters that Anthropic had set a boundary of not running clinical trials for now.^[21]

Reuters also reported, citing one of its sources, that Anthropic wanted to push how far Claude could direct robotic units to carry out experiments with limited human intervention; a spokesperson said human oversight and involvement were essential for safety, and Kauderer-Abrams said the company was "in the very early innings of using AI to automate the execution of lab work."^[21] Anthropic's own description of its lab, published on September 23, was more specific: the Bay Area lab looks like a typical molecular biology lab, works only at biosafety levels 1 and 2, does not handle pathogens that can infect humans, and has all of its lab work performed by human scientists. The post said Anthropic had experimented with AI-accelerated lab work through initiatives such as the Model Hardware Standard but found that approach less suited to the ad hoc workflows of its molecular biology research.^[22] CEO Dario Amodei wrote the same day that it may eventually be possible for Claude to perform experiments by autonomously controlling lab equipment, "but we aren't doing that today."^[24]

Research group and ART discovery (September 2026)

On September 23, 2026, Anthropic introduced "a new life sciences research group and laboratory," formed in spring 2026 and focused on fundamental biology: using Claude to search DNA datasets for uncharacterized protein families, generate hypotheses at scale and test them in the lab.^[22] The team describes itself as scientists with backgrounds in computational discovery of unusual proteins, CRISPR evolution and gene-therapy enzymes, and sits in Anthropic's life sciences organization alongside teams working on drug discovery and on training Claude in biology and chemistry. In a typical survey, Claude reads the literature on a protein family, reproduces established results from public data to check its methods, searches for members or genomic neighbors that fit no described system, writes a short report per candidate, and then critically reviews the evidence; most candidates are eliminated. Surviving candidates are expressed and characterized in the lab. The team works in Claude Science and Claude Code, sometimes with an in-house harness that coordinates many Claude sessions in parallel.^[22]

The first reported result is the array-associated reverse transcriptase (ART) family, described in a technical report whose six Anthropic authors include Kauderer-Abrams.^[23] Agents running Claude Mythos 5 surveyed reverse transcriptases across 1.9 billion metagenomic protein clusters. Anthropic's post summarizes the run as roughly 950 agents using 210 million tokens over 21 hours, gathering more than 200,000 reverse transcriptases and narrowing 3,500 candidate systems to 20 reports; the report gives 949 agent sessions, 215.6 million tokens and 21.5 hours without human intervention.^[22][23] One agent reading raw DNA beside a jumbo-phage reverse transcriptase noticed a tandem repeat array. ART combines the enzyme, a partner gene and a long array of evenly spaced repeats reminiscent of a CRISPR array, and Anthropic's first experiments showed the array is expressed as distinct short RNAs. The enzyme itself had been identified before; Anthropic says Claude appears to be the first to notice the array and partner protein.^[22] The report states that the team has not shown the enzyme is active, and that none of ten repeat runs of the campaign read the DNA upstream of the enzyme.^[23] Feng Zhang of MIT and the Broad Institute, quoted by Anthropic after reviewing the preprint, called the identification "genuinely intriguing" and said it "merits further investigation."^[22] Amodei wrote that its "precise function, biotechnological utility (if any), or level of significance is not yet clear," and noted that a Stanford team had independently described an RT system with a non-coding array that is "in some ways similar."^[24][25]

The post closed with a call for collaborators: Anthropic said it wants to work with other scientists to extend AI-driven hypothesis generation to a broad range of problems in genomics and other fields, and invited scientists with a proposal for a research question to get in touch.^[22]

Reception

Coverage treated the launch as Anthropic's entry into a large vertical, with CNBC, SiliconANGLE, and trade outlets framing it as the company's first formal move into life sciences.^[2][6] Reporting noted that several large pharmaceutical companies were already using Claude. Anthropic cited Sanofi and Novo Nordisk as customers at launch, and later coverage listed Sanofi, Novo Nordisk, and AbbVie among pharma users of Claude.^[5][11]

Anthropic deepened the commitment the following spring. In early April 2026, multiple outlets reported that the company was acquiring Coefficient Bio, a small New York startup whose founders previously worked on machine learning at Genentech, in a stock deal valued at about $400 million, folding the team into its healthcare work to push further into drug discovery.^[11][12]

On September 16, 2026, Novo Nordisk and Anthropic announced a collaboration under which the Danish drugmaker will use Anthropic models and test Claude Science in specific research and development workflows, with the two companies jointly addressing drug discovery challenges identified by Novo's scientists.^[27] Coverage of the September 23 enzyme result stressed its preliminary status: The Next Web noted that the preprint had not been peer reviewed and that Anthropic does not yet know what the system does, while TechCrunch wrote that it would be up to the broader research community to validate how big, or new, the discovery is.^[26][25]

References

1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12Claude for Life Sciences - Anthropic
2. ^1 ^2 ^3 ^4 ^5 ^6Anthropic launches Claude Life Sciences to give researchers an AI efficiency boost - CNBC
3. ^Anthropic Launches Claude for Financial Services - Finovate
4. ^Anthropic announces Claude Life Sciences - Techmeme
5. ^1 ^2 ^3 ^4 ^5 ^6 ^7Anthropic Launches Claude for Life Sciences with Benchling, PubMed Integration - Maginative
6. ^1 ^2Anthropic takes aim at biotech with Claude for Life Sciences - SiliconANGLE
7. ^Life sciences - Claude by Anthropic
8. ^1 ^2Benchling Partners with Anthropic to Build a Bridge Between Science and AI - Benchling
9. ^Advancing Claude in healthcare and the life sciences - Anthropic
10. ^Anthropic rolls out new healthcare and life science features for Claude - Fortune
11. ^1 ^2AI Giant Anthropic Leans Into Life Sciences With $400M Coefficient Bio Catch - BioSpace
12. ^Anthropic buys biotech startup Coefficient Bio in $400M deal: Reports - TechCrunch
13. ^1 ^2 ^3 ^4 ^5 ^6How Claude is uplifting biomolecular modeling - Anthropic, September 17, 2026
14. ^1 ^2 ^3 ^4 ^5Anthropic x Adaptyv Protein Design Competition - Proteinbase
15. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8Introducing the Life Sciences Verification Program - Anthropic, September 17, 2026
16. ^1 ^2 ^3Anthropic is operating a lab that conducts biology experiments - Julie Bort, TechCrunch, September 18, 2026
17. ^Claude Opus 5.5 - Anthropic, September 22, 2026
18. ^How Claude is accelerating protein design and analytical chemistry - Anthropic, August 18, 2026
19. ^Case study: Benchmarking Claude's protein designs in the wet lab - Adaptyv Bio, August 20, 2026
20. ^1 ^2Claude for Financial Services - Anthropic, July 15, 2025
21. ^1 ^2 ^3 ^4 ^5Exclusive-Anthropic quietly sets up biology lab as it ramps AI drug program - Jeffrey Dastin and Michael Erman, Reuters (via Yahoo Finance), September 18, 2026
22. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8Claude discovers a novel enzyme system with CRISPR-like repeats - Anthropic, September 23, 2026
23. ^1 ^2 ^3Autonomous AI agents discover reverse transcriptases with tandem repeat arrays - Peter H. Yoon, Januka S. Athukoralage, Emmanuel Ameisen, Eric Kauderer-Abrams, Nicholas T. Perry and Matthew G. Durrant, Anthropic technical report, September 2026
24. ^1 ^2Post on X - Dario Amodei, September 23, 2026
25. ^1 ^2Anthropic says its biology lab has already found something big - Julie Bort, TechCrunch, September 23, 2026
26. ^Anthropic says Claude found a new enzyme system with CRISPR-like repeats - Ana Maria Constantin, The Next Web, September 23, 2026
27. ^Novo and Anthropic will collaborate to advance drug discovery with Claude - Novo Nordisk (GlobeNewswire), September 16, 2026