Anthropic has built a physical biology lab: They want to teach Claude to run real experiments

Anthropic — the American AI company that created Claude — has stepped beyond computer experiments and set up its own biology lab in the San Francisco Bay Area. There, the company is already conducting physical research and studying how AI could control lab robots and help scientists test hypotheses not only on screen but in the real world.

The lab's existence was reported on September 18 by Reuters, citing two people familiar with the project. Anthropic's head of Life Sciences, Eric Kauderer-Abrams, confirmed to the agency that the company indeed operates a so-called wet lab — a "wet laboratory" where real biological and chemical experiments are conducted.

This is a notable step for the AI industry. Until now, most AI talk in biology concerned data analysis, protein modeling, searching for potential molecules, and generating scientific hypotheses. Now Anthropic is trying to close the loop: AI proposes or helps plan an experiment, lab equipment executes it, and the results are returned to the system for further analysis.

What is a "wet lab" and why does an AI company need one?

A wet lab is a standard physical laboratory where scientists work with biological samples, reagents, cells, proteins, and lab instruments. It differs from in silico research, which is done entirely on a computer.

For Anthropic, such a lab is needed primarily because a computer model itself can't prove that a biological hypothesis works in reality.

"In biology, the final verification still takes place in a real laboratory and likely will for a long time," Kauderer-Abrams explained in a Reuters interview.

According to him, Anthropic already performs some work at its own facility and outsources some research to external partners — much like ordinary biotech companies do.

However, a company representative clarified an important detail to Reuters: Anthropic's own lab is not intended solely for drug development. The company hasn't yet disclosed the specifics of its research areas.

Claude to be connected to lab robots

One of the most unusual parts of the project is the attempt to let AI interact directly with physical equipment.

According to Reuters, Anthropic is exploring a scenario where Claude could operate robotic lab instruments and carry out experimental procedures with limited human intervention.

A fully autonomous lab doesn't exist yet. Anthropic emphasizes that human oversight remains necessary, including for safety reasons.

But the technological foundation for such a system is already being built.

In August, Anthropic introduced the Model Hardware Standard — a standard that allows AI agents to interact with various physical devices. In a research lab, these could be microscopes, automatic liquid dispensers, robotic manipulators, and other equipment.

The company envisions that instead of separate programs for each piece of equipment, the AI could coordinate multiple devices at once, adjust experiment parameters on the fly, and respond to certain hardware errors.

Anthropic is already testing this technology with scientific organizations and companies involved in lab automation.

How is this fundamentally different from ordinary Claude use?

Today, a researcher can ask an AI to review scientific literature, process data, write code, or suggest possible explanations for experimental results.

But between a good hypothesis and a scientific discovery lies the most critical step — physical verification.

For instance, the AI might hypothesize that a certain molecule affects a target protein. To find out if it actually happens, the substance must be synthesized or obtained, an experiment must be run, the result measured, and then repeated.

If software, AI, and lab robotics can be tied into a single system, part of that cycle could theoretically be performed automatically and around the clock.

This potentially shortens the time between a hypothesis's appearance and its experimental testing — one of the main bottlenecks in modern biology.

Anthropic is moving more actively into medicine and biology

The physical lab didn't appear as an isolated experiment. Over the past year, Anthropic has been systematically expanding its presence in life sciences.

In June, the company launched Claude Science — a specialized working environment for scientists that combines scientific databases, computational tools, and Claude's capabilities.

Anthropic also collaborates with major research and pharmaceutical organizations, including the Howard Hughes Medical Institute, Allen Institute, Genentech, Bristol Myers Squibb, and Novo Nordisk.

The company provides researchers with computational resources and grants, including for projects focused on rare genetic diseases.

And on September 17 — a day before the news of the physical lab broke — Anthropic opened applications for its Life Sciences Verification Program. The program gives vetted biology and pharmaceutical specialists access to more powerful models with less restrictive limits on professional biological queries.

The company explains that many legitimate scientific tasks might superficially resemble queries that safety systems block due to misuse risk.

Anthropic also wants to work on early-stage drug development

Earlier in the summer, the company signaled ambitions beyond making software for pharma companies.

At an Anthropic event in San Francisco, Kauderer-Abrams spoke about the possibility of running its own preclinical research programs, especially for diseases that the traditional pharmaceutical industry considers commercially unattractive.

The company bets that AI can speed up the search for approaches to diseases long considered too difficult for drug development.

In particular, Anthropic is interested in complex biological molecules, including bispecific and trispecific antibodies that can interact with multiple targets at once.

However, there's a huge distance between finding a promising molecule and a new drug reaching the market. Most candidates fail during preclinical studies and safety and efficacy trials.

Anthropic doesn't plan to conduct human clinical trials yet.

Kauderer-Abrams explained this is partly because the company doesn't want to become a direct competitor to pharmaceutical and biotech companies that use its AI tools.

Why such a lab has a darker side

The better AI gets at biology, the harder the safety question becomes.

The same model could potentially help scientists find drugs while also providing knowledge that could be used in dangerous biological experiments.

This issue is especially relevant for Anthropic. In September, the company itself published a report on Claude misuse cases, describing several studies with potential applications in bioweapons development. Anthropic emphasized that it couldn't always determine the real intent of users.

The company blocked the relevant activity and continues to tighten safeguards for the riskiest requests.

Against this backdrop, launching its own physical lab looks dual-edged: Anthropic simultaneously tries to make Claude more useful for biologists while preventing the same capabilities from lowering the barrier to dangerous experiments.

According to Kauderer-Abrams, finding the balance between accelerating drug development and limiting risks remains a key principle of the Life Sciences division.

Why the physical lab changes AI's role

Until now, generative AI has mostly remained a program that writes text, analyzes documents, creates images, or helps with code.

Pairing AI with a robotic lab alters the nature of the technology.

The model gains the ability not only to reason about the physical world but to take part in a real experimental cycle: formulate a task, send commands to equipment, receive measurements, analyze the result, and propose the next step.

For now, Anthropic is only at an early stage of this process. The company doesn't disclose either the lab's size, the project's cost, or the list of experiments already being performed inside.

It's also unknown how autonomously Claude can currently work with real equipment. Reuters only reports Anthropic's intention to develop this direction, and the company itself emphasizes the need for human involvement.

But the very existence of its own biology lab shows where the biggest AI developers are heading: from chatbots and computer models to systems that can interact with real scientific equipment.

If this approach proves effective, one of the main fields of competition between AI companies may no longer be just answer quality, but the speed at which they can turn digital hypotheses into testable physical experiments.

Sources: Reuters, Anthropic — Model Hardware Standard, Anthropic — Life Sciences Verification Program.