Anthropic has established a wet biology laboratory in the San Francisco Bay Area as it expands efforts to use AI in drug discovery, according to Reuters. The company has said its life-sciences work is aimed at rare and neglected diseases and is focused on preclinical research rather than running clinical trials. The laboratory gives Anthropic a physical environment in which researchers can test hypotheses generated with the help of Claude and connect computational work with conventional biological experiments.
The move expands Anthropic’s activities beyond model development. Most AI companies work with software, data and computing infrastructure, but drug discovery requires experiments in the physical world. A model can propose a molecular structure or analyze a biological dataset, but those predictions still have to be tested in a laboratory to determine whether they correspond to real biological effects.
Reuters reported that Anthropic’s life-sciences team is led by Eric Kauderer-Abrams and that the company is using Claude to automate scientific tasks. The company is also expanding its software for scientific work. The goal is not to replace laboratory research but to accelerate parts of the process that involve reading papers, analyzing data, generating hypotheses and planning experiments.
Why AI needs a physical laboratory
Drug discovery is particularly suited to this kind of workflow because researchers face large volumes of scientific literature and experimental data. An AI system can search and summarize information much faster than a human can read every paper, but it can also make mistakes or combine evidence incorrectly. Human researchers So remain important for deciding which hypotheses are plausible and which experiments are safe and informative.
Anthropic’s decision to build a physical lab also shows the limitations of purely software-based AI research. Biology contains variables that are difficult to represent perfectly in training data. Cells can respond differently depending on experimental conditions, and biological systems contain complex interactions that are not captured by a single measurement. Laboratory experiments provide the feedback needed to determine whether a computational prediction works in practice.
The company is focusing on preclinical work, meaning research that takes place before human clinical trials. That includes identifying potential treatments and testing their biological effects. Clinical development involves additional regulatory requirements, patient safety considerations and large-scale trials. Anthropic is So not presenting its lab as a replacement for the pharmaceutical development process.
From model output to experimental evidence
Anthropic’s work also includes partnerships with pharmaceutical companies and the acquisition of Coefficient Bio, according to Reuters. Those relationships give the company access to domain expertise and data that can complement its model development. The combination of AI systems, scientific software and physical experiments could allow Anthropic to build a feedback loop in which models suggest experiments and experimental results improve future analysis.
There are significant safety considerations. AI systems that can reason about biology can potentially accelerate useful research, but the same capabilities could be misused. Anthropic has historically emphasized safeguards around powerful models, and its decision to keep humans involved in the laboratory process provides one layer of oversight. The company still has to control what models can access, what actions they can take and how biological information is handled.
Privacy is another issue because pharmaceutical research can involve proprietary compounds, unpublished results and patient-related information. Companies using AI in drug discovery need to understand whether data is retained, how it is isolated and who can access model outputs. Anthropic’s enterprise customers will So evaluate the life-sciences tools partly on the same security and data-governance criteria used for other business AI systems.
- Literature analysis and scientific information retrieval
- Hypothesis generation and experiment planning
- Preclinical biological testing
- Human review of experimental results and safety decisions
The investment in biology is also commercially significant. AI companies are looking for new markets where their models can create measurable value, and scientific research provides workflows that can benefit from better reasoning and information retrieval. Drug discovery is expensive and slow, so even a modest reduction in the time required for early-stage research could have practical value if the resulting hypotheses are reliable.
Anthropic is not alone in applying AI to science. Google, Microsoft, Nvidia and specialized startups are developing systems for protein structure prediction, scientific literature analysis, molecular design and laboratory automation. The competition means Anthropic will need to show that Claude-based workflows produce useful experimental results rather than simply attractive demonstrations.
The new biology lab is So best understood as an extension of Anthropic’s model-development strategy into the physical sciences. The company is testing whether AI can participate in a complete research loop that includes information gathering, hypothesis generation, experiment planning and measurement. The results will depend on the quality of both the AI system and the scientists who decide which predictions are worth testing.
Source: https://www.reuters.com/world/anthropic-quietly-sets-up-biology-lab-it-ramps-ai-drug-program-2026-09-18/.
The laboratory approach also gives Anthropic a way to evaluate whether AI-generated scientific work survives contact with experimental reality. A useful model output is not necessarily a useful scientific result. Researchers still have to reproduce an observation, measure its effect and determine whether the result is statistically and biologically meaningful. That feedback can reveal weaknesses in the model’s assumptions that are invisible in text-only evaluation.
Preclinical research is also where safety controls become important. The company can restrict which systems are connected to laboratory equipment and require human approval before experiments are executed. Separating model-generated recommendations from physical actions allows researchers to benefit from AI assistance without giving the model unrestricted control over the laboratory environment.
Anthropic has confirmed that it operates a wet biology laboratory and is using its AI systems in physical experiments, Reuters and TechCrunch reported on September 18. The company acquired Coefficient Bio in April and has been expanding its life-sciences work. Anthropic’s head of life sciences, Eric Kauderer-Abrams, told Reuters that final biological validation still requires real laboratory work. The lab So represents a move from AI-assisted analysis toward AI-supported experimentation.
Anthropic’s broader life-sciences strategy is also visible in its September 17 Life Sciences Verification Program. The company says the beta program gives qualified life-science professionals access to Mythos, Opus and Sonnet models with safeguards adjusted for biology-related work. Anthropic says dozens of organizations were already onboarded through early access and that the program covers areas including drug discovery, research biology, clinical development and manufacturing.
The laboratory matters because biological research has a physical feedback loop. A model can propose a molecular target, analyze a dataset or suggest an experimental design, but the proposed result still has to be observed in a real system. That creates opportunities for AI to accelerate iteration while also creating safety requirements around what an automated system can execute. Human researchers remain responsible for interpreting experimental results and deciding which procedures should proceed.
Anthropic’s approach is different from simply adding a biology chatbot to Claude. A physical laboratory can provide proprietary experimental data and a controlled environment in which AI-generated hypotheses are tested. Over time, that could allow the company to study where language-model reasoning helps and where biological reality exposes gaps in the model’s assumptions. The scientific value will depend on reproducibility, independent validation and clear separation between model recommendations and laboratory authorization.
Featured image source: VOH image asset.
The new laboratory capability also gives Anthropic a way to connect its model-evaluation work with a domain where mistakes can have physical consequences. Biology models can be tested not only for factual accuracy, but for whether proposed experimental steps are appropriate, reproducible and within defined safety boundaries. That makes the company’s biology work relevant to both scientific productivity and the broader question of how AI systems should be governed when they can influence real-world experiments.