Anthropic Establishes Wet Lab in Bay Area to Advance Real Experiments
Anthropic has set up a wet lab in the San Francisco Bay Area, advancing the biological analysis of Claude from computer-based to real physical experiments. Eric Kauderer-Abrams, head of life sciences, confirmed to Reuters that the company is conducting experiments in its own facilities while also outsourcing some work to external partners, stating that the ultimate validation of biology remains real laboratory experiments for a considerable time.
A spokesperson emphasized that the lab is not specifically for drug discovery and declined to disclose details about its scale, staffing, activation time, and biosafety level. The company continues to outsource some biological work, with its own lab serving as an addition rather than a replacement. Kauderer-Abrams noted that automating experiments with AI is still in its early stages, and human oversight remains essential for safety.
The next step is to have Claude control equipment such as robotic arms, liquid handling robots, and plate readers. The company launched a model hardware standard research preview last month, allowing unified control of microscopes, liquid handlers, and robotic arms, enabling models to adjust parameters based on results and recover autonomously from certain failures. Genentech has already linked liquid handling, robotic arms, and plate readers for protein quantification using this standard; Carnegie Mellon has increased the speed of dose-response experiments by about three times using it.
Anthropic plans to conduct preclinical research focusing on diseases that traditional pharmaceutical companies consider too difficult or lacking commercial value, but will not enter human clinical trials to avoid direct competition with clients like Roche, Genentech, Bristol Myers Squibb, and Novo Nordisk. The company has added Novartis CEO Vas Narasimhan to its board and acquired biotechnology startup Coefficient Bio for approximately $400 million in stock.
Launched in June this year, Claude Science integrates a repository of papers, code, and computational resources into a single workspace, connecting over sixty genomic, protein structure, and cheminformatics interfaces. Life sciences is now one of the company's largest areas of human and resource investment, with ongoing recruitment for procurement leads and biochemists.
In market mechanisms, this represents a model company extending its moat of computational power into a closed loop of wet lab data: what is sold is pure software analysis subscriptions, while what is bought is a platform capable of controlling instruments and generating proprietary experimental data. Funding is directed towards lab automation, preclinical outsourcing networks, and binding workflows of pharmaceutical companies; beneficiaries are instrument vendors and large pharmaceutical clients that have adopted the model hardware standard, while those under pressure are AI drug startups that only perform computational design without feedback from wet experiments.
The company has not disclosed the lab's budget but has provided funding for up to fifty projects in academia, with a maximum of $30,000 in credits for each, directing external experimental traffic into the same toolchain.
Source: Public Information
ABAB AI Insight
Anthropic's entry into life sciences does not start by opening a new biotechnology company, but rather by selling workstations before buying a closed loop. Eric Kauderer-Abrams, after leaving a diagnostic startup, has positioned Claude Science as a research agency comparable to Claude Code, and then acquired Coefficient Bio to fill in target selection and pipeline planning, while placing the head of Novartis on the board to gain entry into the pharmaceutical industry. The narrative of rare diseases, compounded by the death of Dalio Amodei's father before the therapy emerged, carries both product positioning and safety narratives: conducting preclinical studies without human trials avoids client conflicts and the highest regulatory risks.
Capital is being allocated along three lines: first, the wet lab and procurement structure, turning proprietary experiments into private feedback for model training; second, the model hardware standard, writing pipettes, robotic arms, and microscopes into a unified read-write interface to lower replacement costs for pharmaceutical companies; third, selling positions to Genentech, Bristol Myers Squibb, and Novo Nordisk, increasing switching costs with client data and workflows. The motivation is to lock in the "analysis-experiment-parameter adjustment" protocol before similar research agencies like OpenAI.
A similar path is NVIDIA using Bionemo to integrate computational biology models into cloud toolkits, and instrument vendors like Danaher and Eppendorf relying on automation platforms for experimental consumables. The industry is transitioning from software analysis to controlling experimental execution: whoever first establishes a hardware standard can transform laboratories from manual surfaces into callable APIs.
This represents a technological substitution. The mechanism is that the bottleneck of wet experiments lies not in ideas but in fragmented instrument interfaces; once unified, models can write failure samples back into the next round of parameters for the first time. As the execution rights of experiments shift from graduate students on night shifts to agents plus robotic arms, pricing power shifts from the speed of papers to who owns a repeatable physical closed loop.
ABAB News · Cognitive Law
For models to become stronger, first write the laboratory as an interface.
Not conducting human trials is to continue selling to pharmaceutical companies.
A computational closed loop is not enough; wet data is the moat.