Anthropic’s wet lab: why Claude is conducting physical experiments

Anthropic has confirmed that it operates a biological laboratory in the San Francisco Bay Area. The company is already conducting physical experiments there while also working with external partners. The main focus is fundamental biology rather than drug development. For scientific teams, this is an important shift: AI is moving beyond the analysis of text and code.

What has been confirmed

According to Eric Kauderer-Abrams, head of life sciences, the ultimate test in biology remains real laboratory work. The company conducts some research at its own facility and some with partners. In April, Anthropic acquired the biotech startup Coefficient Bio, then launched the Claude Science platform and a validation program for approved researchers.

A company representative clarified that the laboratory was not created specifically for drug development. However, Anthropic has not disclosed the facility’s size, staffing, opening date, or specific projects.

Why this matters for science

A model can analyze literature and data, but a biological result requires an experiment. An in-house laboratory closes the loop: hypothesis, experiment, observation, refinement of the hypothesis. For rare diseases and areas that the pharmaceutical industry is reluctant to pursue, this loop could accelerate the testing of ideas.

However, physical experiments introduce new risks: safety, reproducibility, materials tracking, access control, and protocol accuracy. That is why the company emphasizes the early stage of automation and mandatory human involvement.

Partnership with Novo Nordisk

On 16 September 2026, Novo Nordisk and Anthropic announced a collaboration to accelerate drug development and AI-assisted engineering work. Novo Nordisk plans to test Claude Science in selected research processes and use the models for software engineering. The companies said they would maintain strict data governance and human oversight.

This case illustrates a likely model for deployment: Anthropic’s laboratory tests fundamental processes, while a pharmaceutical partner applies the models to its own data and processes under its own compliance rules.

Where the boundaries lie

The company does not plan to conduct clinical trials. Laboratory-operations automation is at an early stage, and human involvement remains a mandatory safety requirement. This means the article should not be read as an announcement of an autonomous scientist: the goal is to gradually reduce manual workload, not to hand responsibility over to a machine.

What scientific teams should do

• Connect the model to primary data and laboratory notebooks, not just to the text of papers.
• Record the hypothesis, protocol, raw data, and model version before beginning the experiment.
• Separate data analysis, experiment planning, and execution.
• Introduce independent verification of critical conclusions and external audits of protocols.
• Start with reproducible tasks rather than the most ambitious hypotheses.

Practical takeaway

An in-house biological laboratory turns Claude from a text assistant into a participant in the full research cycle. The teams that benefit are those that prepare their data, protocols, and access controls in advance: infrastructure is what determines whether AI becomes an accelerator or a source of irreproducible results.