Anthropic Quietly Builds a Wet Lab as Its AI Ambitions Move Into Drug Science

Anthropic has quietly set up a wet lab, a facility for physical biological experiments rather than computer modeling, in the San Francisco Bay Area, as the company pushes its artificial intelligence work deeper into drug science. Reuters reported the lab's existence on September 18, citing two people familiar with the matter, and the report says Anthropic's head of life sciences, Eric Kauderer-Abrams, confirmed it in an interview on Tuesday, September 15. The lab is up and running.

Related reading: Anthropic Says Claude Discovered a CRISPR-Like Enzyme System in Phage DNA · Anthropic's Claude Duet Works in the Background, and Asks Before It Acts.

The confirmation is a first for Anthropic, which until now had described its biology ambitions mostly in terms of models and data. A physical lab means the company is generating its own experimental evidence rather than reading only the evidence other people produced.

A wet lab, said on the record

A wet lab is the ordinary kind of biology lab, where samples are pipetted, grown and measured at a bench. It is the counterpart to dry-lab work, in which researchers run computations against existing datasets. Reuters reported that Anthropic's turn to physical work marks a step beyond "in silico" evaluations, the computer-only studies that have been the norm for AI biology efforts.

Anthropic says it wants to unlock treatments for rare diseases. Kauderer-Abrams told Reuters that real laboratory work remains necessary to test biological ideas, and a company spokesperson said the facility was not set up specifically for drug discovery, according to reporting that cited Quartz. That distinction matters: a research lab and a drug-development program are different undertakings, and Anthropic has described the former.

Why crossing into physical experiments changes the story

For most of the past decade, AI's contribution to biology has been computational. Models such as DeepMind's AlphaFold predicted protein structures, and follow-up tools scored genetic variants or proposed molecules. Those systems work on data that already exists. A wet lab lets a company generate new data, close the loop between a prediction and a measurement, and test whether a model's suggestion survives contact with a living system.

That loop is the point. A model can rank a thousand candidate molecules, but in a wet lab each candidate still has to be synthesized or ordered and then assayed, step by step, until one shows the desired behavior. Automating parts of that process is where AI companies see a chance to compress timelines, and it is also where the hard operational problems sit: robotic systems are expensive, finicky and hard to adapt to novel protocols.

Anthropic is exploring whether its Claude models can direct robotic units to carry out science experiments with limited human intervention, according to the Reuters report, which cited one person. The company says human oversight and involvement remain essential for safety. "We're in the very early innings of using AI to automate the execution of lab work," Kauderer-Abrams told Reuters, describing an area that "has the potential to bring about meaningful acceleration in so many different processes."

That ambition runs into the messy reality of laboratory science. Biological experiments are sensitive to temperature, timing, contamination and operator judgment, and protocols often change from one week to the next. A robotic system that executes a fixed sequence is useful; one that adapts when an assay fails is much harder to build. The value of automation therefore depends less on whether a model can issue commands than on whether it can notice that something has gone wrong and respond. That is also where safety questions concentrate, which helps explain why Anthropic keeps returning to human oversight in its public statements.

The diseases Anthropic says it wants to target

Kauderer-Abrams' public argument is that AI could fast-track development for conditions previously thought "undruggable," meaning too difficult to treat with conventional medicine. One example he gave is antibody engineering. Bispecific and trispecific antibodies are designed to bind more than one target at once, which opens treatment routes that a single-target drug cannot take, and they are complex enough that design and screening are natural candidates for machine assistance.

The precise diseases Anthropic plans to study, and how far its internal work has progressed, remain unclear. At a June gathering in San Francisco, Kauderer-Abrams said the company would focus on early-stage research in areas large drugmakers have largely overlooked because they offer little financial incentive, according to the Reuters report. That is a familiar gap in pharmaceutical research: rare diseases and some infectious threats attract less commercial investment because the patient populations, and the potential returns, are small.

From a molecule to a medicine is still a long road

AI can accelerate the discovery end of drug development. It does not shorten the regulatory end by itself. Once a promising molecule is identified, it can take years to reach the market through preclinical testing, human trials and regulatory review. Anthropic has not taken on clinical trials, and Kauderer-Abrams told Reuters the company is not competing with pharmaceutical and biotech companies whose business is bringing drugs to patients.

The comparison that hangs over the field is Isomorphic Labs, the drug-discovery arm of Alphabet. It has worked for years toward reaching the clinic, and earlier delayed that goal to the end of 2026. The gap between a clever model and a tested therapy is where most AI drug programs have spent their time, and most drugs that enter clinical trials still fail there.

A broader biology program, and a separate claim

The wet lab is one piece of a larger life-sciences effort at Anthropic. The company launched Claude Science, a research workbench that connects to more than 60 scientific databases across genomics, proteomics and structural biology. It has acquired the startup Coefficient Bio, reported at about $400 million in stock, and works with pharmaceutical companies including Genentech, Bristol Myers Squibb and Novo Nordisk. Life sciences has become one of the company's most heavily staffed units, and Anthropic added Novartis chief executive Vas Narasimhan to its board.

There is also a separate set of numbers that has circulated alongside the lab news, and it should not be confused with it. In a company claim reported by the Associated Press and summarized in a TechStartups roundup, Anthropic said Claude led about 26 percent of its model research and development work by August 2026, up from roughly zero in February. The company said about 90 percent of that R&D involves some form of collaboration with Claude, a broader category that includes tasks done under close human direction. Anthropic says its models are not autonomously designing and training their successors. Those figures describe work on AI models, not work in the biology lab, and they remain the company's own accounting.

The timing: ambition meets anxiety

The disclosure lands in an unsettled moment for the industry, and for Anthropic in particular. The company has spent the past year arguing that frontier AI carries serious risks even as it builds more capable systems. Chief executive Dario Amodei published an essay, "We Must Pace the Frontier," calling for a slower, more deliberate approach to capability gains, and the company has warned that advanced models could be misused in ways that include developing biological weapons. A company opening a biology lab while warning about biological risk is not a contradiction so much as a reflection of how the same tools cut two ways.

Amodei's interest in disease is personal. In a recent essay, he wrote that a disease killed his father only a few years before a cure came about. That experience sits behind Anthropic's stated aim of tackling conditions the pharmaceutical industry neglects, and behind the company's argument that AI investment has to produce broad benefits before staff and the public lose faith that the disruption is worth it.

The broader context is public anxiety about AI's direction. A wave of slowdown commentary, pitched against rivals who argue that heavy restrictions would hand an advantage to China, has made the pace of development a live political question. For a company that has positioned itself as safety-conscious, a physical lab is both a scientific commitment and a message: the work is moving forward.

What to watch

Three things would clarify how much substance sits behind the announcement. The first is whether Anthropic publishes experimental results, which would distinguish a working lab from a well-equipped one. The second is whether its automation experiments produce peer-reviewed evidence that a model can reliably direct robotic equipment. The third is how the company's neglected-disease focus interacts with its commercial partnerships, since the diseases it says it wants to chase are the ones with the least obvious path to revenue.

For now, the confirmed fact is narrow and clear. Anthropic has a biology lab, it has said so on the record, and it is running experiments. Whether that shortens the years-long road between a molecule and a medicine is a question the lab itself has yet to answer.

The Anthropic wordmark (CC0 via Wikimedia Commons) on a TechPastWeek-original background.