Anthropic has confirmed it is operating a physical wet-biology lab in the San Francisco Bay Area, moving the Claude AI maker beyond spreadsheets and simulations into hands-on experimental science staffed by human researchers rather than autonomous robots. The confirmation came on September 18, 2026, when Anthropic’s head of life sciences, Eric Kauderer-Abrams, addressed the lab directly in a Reuters interview, with coverage syndicated across outlets including SiliconANGLE, The Next Web, and Geo News, and TechCrunch independently reporting that same day that the lab was using Claude models to help run physical biology experiments, making it one of the most discussed AI stories of the week.
The move lands two days after Anthropic and Danish drugmaker Novo Nordisk announced a separate collaboration to test the company’s Claude Science tools on real drug-discovery problems. Together, the two stories mark a clear pivot: Anthropic, a company built on training and selling large language models, is now pairing Claude’s “biological reasoning” with physical, human-run lab validation to prove those predictions actually hold up outside a simulation.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Anthropic Confirms a Physical Biology Lab in the Bay Area
According to Reuters, two sources familiar with the matter said Anthropic set up the lab to go beyond computer-based simulation work, with early reporting pointing to Claude directing robotic units to carry out biology experiments. The company confirmed the lab exists but, as The Next Web noted in its September 18 coverage, initially declined to disclose the facility’s size, staffing level, or opening date. Anthropic has since filled in part of that picture, saying the lab runs at BSL-1 and BSL-2 biosafety levels, does not handle human-infecting pathogens, and that all lab work is performed by human scientists rather than autonomous robots.
That silence is notable given how much Anthropic has publicized its other 2026 initiatives, from new Claude model releases to its ongoing push for an industry-wide AI safety slowdown. The biology lab news broke without an official Anthropic press release; it surfaced through Reuters sourcing and a company spokesperson’s limited confirmation, which is why outlets have repeatedly used the word “quietly” to describe the lab’s construction.
SiliconANGLE’s report on the opening initially framed the effort around Claude-directed lab automation, but Anthropic’s subsequent clarification puts trained human scientists at the bench, with Claude Science supporting protocol planning and data analysis rather than operating the equipment itself. No specific robotics vendor, instrument manufacturer, or automation platform has been named in any of the reporting so far.
Why a Chatbot Company Needed a Wet Lab
Anthropic’s core business is renting access to Claude models through APIs and subscriptions, not running biotech facilities. So why build a wet lab at all? The answer, according to the company’s head of life sciences, comes down to a limitation that no amount of additional training data can fix: biology does not fully validate on a screen.
Anthropic has spent much of 2026 expanding Claude Science, a platform aimed at researchers who want to use Claude for literature review, hypothesis generation, and computational modeling. Reports say Anthropic used Claude to optimize more than 30 open-source biomolecular-modeling systems in under four weeks, delivering roughly a fourfold average speed improvement. That is meaningful progress, but it is still confined to code and simulation. A model can predict how a protein might fold or how a molecule might bind a target, but the only way to know for certain is to run the reaction in a physical lab and measure the result.
That gap between simulated and physical validation is exactly what Anthropic’s new facility is built to close. It also fits a broader pattern the company has been signaling since June 2026, when it first flagged plans to automate science more broadly, starting with Claude Science, according to Anthropic’s own Claude for Life Sciences announcement.
Eric Kauderer-Abrams: Real Lab Work Is “the Final Test”
The most detailed public comments on the project come from Eric Kauderer-Abrams, Anthropic’s head of life sciences, in a Reuters interview published September 18, 2026 and mirrored on Yahoo Finance. Kauderer-Abrams framed the physical lab as a necessary complement to computational work rather than a replacement for it.
“We believe that to do biology, the final test is still and will be for a while in real lab work.”
Eric Kauderer-Abrams, Head of Life Sciences, Anthropic — via Reuters
He also addressed the balance between in-house facilities and outside partners directly, describing Anthropic’s approach as consistent with how the broader biotech industry already operates rather than something radically new.
“We absolutely are doing that today, and I would describe our approach as being typical of what you would see in most biotech companies, where there’s some amount of that that we’re doing in our own facilities and some amount of that that we’re working with external partners on.”
Eric Kauderer-Abrams, Head of Life Sciences, Anthropic — via Reuters
Kauderer-Abrams also acknowledged the tension between moving fast and moving carefully, saying the company is “always balancing these things,” and described the broader mission as making models better in ways that will “uplift and accelerate developing medicines” while being deliberate about deployment to limit risk. That cautious framing echoes Anthropic’s public posture on AI safety more broadly, a theme the company has pressed hard this year, including in its push for a three-step AI slowdown proposal from CEO Dario Amodei.
Anthropic’s Biology and Pharma Moves in 2026
The wet lab did not appear in isolation. It is the latest in a string of life-sciences moves Anthropic has made through the year, several of which converged in the same week of September 2026.
| Date (2026) | Development | Source |
|---|---|---|
| June | Anthropic signals intent to automate science more broadly, launching early Claude Science work | R&D World |
| September 16 | Novo Nordisk and Anthropic announce collaboration to test Claude Science on drug-discovery workflows | Reuters |
| September 16 | Anthropic says goal is for Claude to support “the entire process, from early discovery through to translation and commercialization” | Anthropic official announcement |
| Reported by mid-September | Anthropic and Adaptyv Bio run a protein-design competition covering wet-lab validation of more than 5,000 designs | Brussels Morning |
| September 18 | Reuters reveals Anthropic has quietly built an in-house wet lab in the Bay Area | Reuters |
| September 18 | Eric Kauderer-Abrams confirms the lab exists in a Reuters interview, says it is “not for drug discovery specifically” | Reuters, The Next Web |
| Following the Sept. 18 report | Anthropic confirms the lab operates at BSL-1 and BSL-2, handles no human-infecting pathogens, and is staffed entirely by human scientists | Anthropic |
Taken together, these moves show a company trying to build credibility across the entire drug-development pipeline: computational modeling through Claude Science, partnership-based testing with a major pharma company in Novo Nordisk, competitive validation through the Adaptyv Bio protein-design effort, and now direct physical experimentation inside its own walls.
Inside Claude Science: From Code to Cell Cultures
Claude Science is the umbrella under which Anthropic packages its research-oriented tools for scientists, and it is the software layer that connects the company’s language models to laboratory work. Reporting indicates Anthropic used Claude to rewrite and speed up more than 30 open-source biomolecular-modeling tools, cutting runtimes by roughly a factor of four in under a month of work. That kind of tooling optimization is a natural extension of what large language models already do well: reading, refactoring, and improving code.
The harder, less proven step is what the new wet lab is testing: whether Claude can meaningfully plan and guide the design of an experiment, from protocol design to interpreting instrument output, even as physical execution stays in human hands. Early Reuters sourcing described the lab in terms of Claude directing “robotic units,” but Anthropic has since clarified that all lab work is performed by human scientists, not autonomous robots, and that the facility does not handle human-infecting pathogens. SiliconANGLE has speculated that the effort may connect to Anthropic’s separate Model Hardware Standard work, but that link has not been confirmed by Anthropic itself. Alongside the lab disclosure, TechCrunch reported that Anthropic used the same September 18, 2026 window to launch a Life Sciences Verification Program, a vetting process through which the company says approved outside researchers get direct access to its most powerful models for biological work, extending the lab’s hands-on caution into how its software gets distributed as well.
What Reporting Does Not Confirm
It is worth being precise about the limits of what is publicly known. No source has disclosed the lab’s funding, square footage, headcount, or experiment throughput, and no outlet has confirmed that the lab is running high-throughput compound screening, gene editing, or animal studies. What has been confirmed: the facility operates at BSL-1 and BSL-2 biosafety levels, does not handle human-infecting pathogens, and all lab work is carried out by human scientists rather than autonomous robots. Reuters also reported that Anthropic is not running clinical trials for now, and a company spokesperson said the facility is “not for drug discovery specifically,” while declining to explain its precise purpose in more detail. TechCrunch’s own September 18, 2026 reporting reinforced that framing, describing the lab’s work as centered on fundamental biology research rather than drug discovery, echoing Reuters’ account that clinical trials remain off the table for now.
The Novo Nordisk Deal: Claude Meets Drug Discovery
Two days before the wet-lab story broke, Anthropic and Novo Nordisk announced they were expanding an existing relationship to put Claude Science to work on drug-discovery problems chosen directly by Novo’s own scientists and computational teams. Under the arrangement, Novo will test Anthropic’s models on specific research-and-development workflows rather than rolling Claude out across all of its drug-development operations at once, a phased approach that mirrors how most enterprises are adopting frontier AI models in regulated industries.
Novo Nordisk already uses Claude for administrative and documentation-heavy tasks such as clinical trial-report generation, work that reportedly cut some documentation timelines from months to minutes. The new collaboration pushes further into scientific territory, aiming Claude Science at “biological reasoning” problems rather than paperwork. Anthropic’s own framing, published alongside the deal, put the ambition in plain terms.
“Now, our goal is to make Claude capable of supporting the entire process, from early discovery through to translation and commercialization.”
Anthropic, official company statement — Claude for Life Sciences announcement
That is a sweeping goal for a company that, three years ago, was primarily known for chatbot safety research. It also puts Anthropic in more direct competition with specialized biotech AI firms that have spent a decade building tools purpose-built for exactly this kind of pipeline work.
Adaptyv Bio and the 5,000-Protein Experiment
The clearest example so far of Claude’s output being validated in a physical lab comes from a reported collaboration with Adaptyv Bio, a protein-engineering company. According to Brussels Morning’s coverage, Anthropic and Adaptyv Bio ran a protein-design competition that included wet-lab validation of more than 5,000 individual protein designs, a scale that would be extremely difficult to test by hand within a reasonable timeframe.
The available reporting does not specify the protein targets involved, the success rate of the designs, or exactly when the validation work took place. But the scale of the number, more than 5,000 designs pushed through physical testing, illustrates why Anthropic sees value in owning lab infrastructure directly rather than relying solely on external partners for every physical check. Farming out thousands of validation runs to a third party takes time and coordination; running a subset in-house lets Anthropic iterate faster on its own models.
What Anthropic Says the Lab Is Not For
Anthropic has been unusually careful to draw a boundary around what its new facility does not do. Kauderer-Abrams told Reuters the company is not running clinical trials for now, a decision framed internally as a way to avoid direct competition with the drugmakers that are simultaneously becoming its customers. A company spokesperson also told reporters the lab is “not for drug discovery specifically,” even while declining to clarify what its specific purpose is.
That distinction matters commercially. If Anthropic positioned itself as a drug developer competing head-to-head with Novo Nordisk, Bristol Myers Squibb, or Genentech, all three would have far less reason to license Claude Science. By keeping the lab focused on tool validation and model improvement rather than pursuing its own therapeutic pipeline, Anthropic can keep selling shovels to the pharmaceutical industry instead of becoming a rival prospector.
Competitive Landscape: Where OpenAI and Google DeepMind Stand
One of the more striking gaps in the current reporting is the absence of a direct rival announcement. As of this week, there is no confirmed report of OpenAI or Google DeepMind operating an equivalent in-house wet lab staffed to run physical experiments tied to a frontier model’s output. Google DeepMind’s most visible biology win to date remains its protein-structure prediction work, which reshaped computational biology but stopped short of physical lab automation. OpenAI has talked about scientific applications of its models but has not disclosed a comparable physical-lab buildout.
That gives Anthropic a first-mover claim in this specific niche, at least based on public disclosures, even as all three labs continue to compete aggressively on model capability. The competitive dynamic among frontier labs has been especially tense in 2026, with Anthropic, OpenAI, and Google also locked in parallel safety talks involving five major AI labs, even as they race each other on commercial deployments like this one. Anthropic’s own model lineup has kept pace on the benchmark side too, with comparisons such as GPT-6 Astra vs. Gemini 3.8 vs. DeepSeek V4.1 Flash showing how tightly the leading labs are bunched on raw capability, which makes differentiated moves like a physical biology lab more valuable as a point of distinction.
AI-in-Biology Approaches Compared
The table below lays out how Anthropic’s approach compares with the publicly known strategies of its two closest frontier-model rivals, based on current reporting rather than speculation.
| Company | Public biology strategy | Physical wet lab confirmed? | Named pharma partners |
|---|---|---|---|
| Anthropic | Claude Science platform, in-house wet lab staffed by human scientists, protein-design validation, pharma R&D partnerships | Yes (Reuters, Sept. 18, 2026) | Novo Nordisk, Bristol Myers Squibb, Genentech (reported) |
| Google DeepMind | Computational protein-structure prediction and modeling research | Not confirmed in current reporting | Not disclosed in this reporting cycle |
| OpenAI | General-purpose scientific reasoning and research assistance | Not confirmed in current reporting | Not disclosed in this reporting cycle |
The gap in the middle column is the story here. Whatever OpenAI or Google DeepMind are doing behind closed doors in biology, neither has a public, sourced report of a company-operated wet lab tied to its own model’s output the way Reuters describes for Anthropic. That could change quickly given how fast this space is moving, but for now it is a distinct edge, or at least a distinct headline, for Anthropic.
Historical Context: From AlphaFold to Autonomous Lab Robots
AI’s role in biology has moved through fairly distinct phases over the last several years. The first wave centered on prediction: models trained to guess how proteins fold or how molecules might interact, all without leaving a computer. That wave delivered genuinely useful science but always required a follow-up step in a physical lab to confirm the predictions actually held up.
A second wave, anticipated by much of the industry commentary around AI and biology, imagined closing that loop entirely inside one organization: a model proposes an experiment, a robot runs it, the model reads the result, and human involvement shrinks at every step. Anthropic’s actual lab is more conservative than that vision. The company says human scientists, not robots, carry out all of the physical work, with Claude Science supporting analysis and experiment design rather than execution. That still echoes the “agentic” pattern that has already reshaped software development, where AI models increasingly write, test, and debug code with limited supervision, but applying even the planning half of that loop to biology is a higher-stakes proposition than code, since a failed prediction in code produces an error message, while a failed prediction in a wet lab can mean wasted reagents, contaminated samples, or, in more sensitive domains, safety concerns that regulators care about.
That is part of why Anthropic’s framing, emphasizing human-run lab work and a boundary against clinical trials, reads as much like risk management as it does product strategy. It also lines up with the company’s broader public posture this year, including its three-step proposal to pace frontier AI development even as it keeps shipping new capabilities.
Market and Industry Impact
For the pharmaceutical industry, the immediate impact is a new option in an already crowded field of AI-for-drug-discovery vendors, but one now backed by an AI lab that also happens to make some of the most widely deployed enterprise language models. Novo Nordisk’s decision to expand its Anthropic relationship into R&D-specific workflows signals that at least one major drugmaker sees enough promise to move past documentation automation and into scientific problem-solving. Reports naming Bristol Myers Squibb and Genentech as existing Claude customers suggest Anthropic is building a roster of pharma relationships that extend beyond a single flagship partnership.
For Anthropic’s own business, the lab is also a credibility play. Selling an AI model on the promise that it can reason about biology is one thing; showing that the same model’s outputs survive contact with a physical lab bench is a different, much harder claim to make, and a much more valuable one if it holds up. That distinction matters to enterprise buyers evaluating whether to trust Claude Science with expensive, time-sensitive research budgets rather than treating it as another chatbot subscription. The stakes behind that credibility play have grown quickly: according to Venture Atlas, Anthropic’s annualized revenue run rate exceeded $65 billion by July 2026, up from roughly $9 billion just seven months earlier in December 2025, a trajectory that hands the company far more capital to fund exactly this kind of physical-lab investment.
There is also a talent and infrastructure signal here. Building and staffing a wet lab, even a modest one, requires biosafety expertise, laboratory equipment procurement, and physical security considerations that are entirely different from the GPU clusters and data centers Anthropic is used to managing. That Anthropic chose to make this investment, quietly, rather than folding biology work entirely into partner facilities, suggests the company expects physical-world experimentation to remain a core part of its roadmap rather than a short-term pilot.
Risks, Oversight, and the Safety Debate
Even with human scientists performing all of the physical work, applying AI reasoning to biology raises a different category of risk than letting a model write code or draft emails. Biology work touches biosecurity concerns that regulators and safety researchers have flagged for years, particularly around dual-use research that could, in theory, be misapplied. Anthropic’s public statements lean heavily on the idea that human oversight remains part of the loop, that the facility is limited to BSL-1 and BSL-2 work with no human-infecting pathogens, and that the lab is deliberately kept away from clinical trials, for now, and, per the company, away from drug discovery specifically.
That said, the company’s own vagueness about the lab’s precise purpose leaves room for scrutiny. Declining to state what a biology lab is for, while confirming that it exists and that Claude Science is being tested as a planning layer around human-run lab work, is the kind of answer that tends to invite more questions rather than fewer, especially from an industry press corps already primed to ask hard questions about frontier AI companies. This is unfolding against the backdrop of Anthropic’s own repeated calls for a slower, more cautious approach to frontier AI development, a stance that has put it at odds with parts of the industry, including reported friction over a proposed AI development slowdown deal. Running a still-opaque physical-science lab while publicly pushing for more caution elsewhere is a contrast worth watching as more details emerge.
What Comes Next: Five Predictions
- More named pharma partners. Following the Novo Nordisk deal and reported Genentech and Bristol Myers Squibb ties, expect Anthropic to announce additional enterprise life-sciences partnerships in the coming months as it builds out Claude Science’s customer base.
- Slow, partial disclosure of the lab’s scope. Anthropic has already filled in some detail directly, confirming BSL-1 and BSL-2 biosafety levels and human-only staffing, but lab size, headcount, and experiment throughput remain undisclosed; expect further reporting rather than a formal Anthropic announcement to fill in the rest.
- Rival labs respond with their own biology plays. With no confirmed equivalent wet lab from OpenAI or Google DeepMind yet, competitive pressure makes it likely one or both will publicize deeper biology-focused initiatives within the next year, even if they stop short of matching a physical lab.
- Regulatory and safety scrutiny increases. As AI models take on more autonomous roles in physical experimentation, expect biosecurity researchers and possibly regulators to ask more pointed questions about oversight structures, especially given Anthropic’s own advocacy for AI safety guardrails elsewhere in its business.
- Anthropic keeps the “not drug discovery” line, but the boundary blurs. As the lab supports more pharma R&D work in practice, expect the distinction between “supporting drug discovery” and “not for drug discovery specifically” to become harder to maintain publicly.
Frequently Asked Questions
Did Anthropic officially announce the biology lab?
Not through a dedicated press release. The lab’s existence surfaced through Reuters reporting published September 18, 2026, citing sources familiar with the matter. Anthropic’s head of life sciences, Eric Kauderer-Abrams, confirmed the lab exists when asked directly in a Reuters interview, but the company did not proactively publicize it, which is why multiple outlets described the buildout as done “quietly.”
Where is Anthropic’s biology lab located?
Reporting places the facility in the San Francisco Bay Area, consistent with Anthropic’s headquarters location. The company has not disclosed a specific city, address, or facility size.
Is Claude actually controlling robots by itself?
No. Early Reuters sourcing raised the possibility of Claude directing robotic lab systems, but Anthropic has since confirmed that all lab work is performed by human scientists, not autonomous robots. Claude Science supports the analysis and planning side of the work, while physical experiments stay in human hands at the bench.
Is this related to the Novo Nordisk partnership?
They are separate but related announcements made within days of each other. The Novo Nordisk deal, announced September 16, 2026, focuses on testing Claude Science against Novo’s own R&D problems. The wet lab, reported September 18, 2026, is an in-house Anthropic facility not tied exclusively to any one partner.
Does Anthropic plan to develop its own drugs?
Current reporting says no, at least for now. Kauderer-Abrams told Reuters that Anthropic is not running clinical trials for now, and a company spokesperson said the lab is “not for drug discovery specifically,” positioning Anthropic as a tool and platform provider rather than a competing drugmaker.
What is Claude Science?
Claude Science is Anthropic’s platform for applying Claude models to scientific research tasks, including literature analysis, hypothesis generation, and computational biology tooling. It is the software layer connecting Anthropic’s language models to the physical lab work described in current reporting.
Are OpenAI or Google DeepMind doing anything similar?
As of this reporting cycle, no source has confirmed an equivalent in-house wet lab at OpenAI or Google DeepMind tied to a frontier model’s biological research output. Google DeepMind’s most prominent public biology work remains centered on computational structure prediction rather than physical lab automation.
What is the Adaptyv Bio connection?
Reports describe a protein-design competition between Anthropic and protein-engineering company Adaptyv Bio that included wet-lab validation of more than 5,000 protein designs, making it one of the clearest examples so far of Claude-generated designs being tested against physical results.
