Biotechnology · global
As Pharma Pulls Back on AI Slogans, Anthropic Chooses to Put Claude Into Experimental Workflows
The point of Claude Science is not to claim that AI can replace drug R&D, but to connect models, databases, programming environments, and traceable records into the same scientific workbench; this also shifts tech companies’ responsibility in biomedical R&D from polished demos toward harder-to-avoid questions of validation.
After the drug discovery boom cooled, AI did not leave the laboratory. It simply moved gradually from center-stage slogans into more practical and more difficult workflows. Anthropic’s launch of Claude Science at this moment, along with its disclosure that it will begin an internal early-stage drug discovery program, shows that large model companies no longer want only to provide chat interfaces or coding assistants, but are trying to take part in how scientific judgments are formed, recorded, and checked.
Anthropic announced on June 30 that Claude Science had opened in beta, positioning it as an AI workbench for scientists. The company described it as its most important expansion in life sciences to date: the platform integrates commonly used research tools, programming packages, scientific databases, auditable outputs, and computing resources in one environment, and is preconfigured for scenarios including genomics, single-cell analysis, proteomics, structural biology, and cheminformatics.
The specific purpose of this type of tool is not to let a model “invent drugs” out of thin air. According to cases disclosed by Anthropic, test users have used Claude Science for single-cell RNA sequencing analysis, CRISPR screen design, protein structure prediction, and cheminformatics analysis; Manifold Bio was also mentioned as having used it to propose targets that could be taken into further experiments. These tasks all sit in the early stages of drug R&D: they can narrow the search space, organize evidence, or generate candidate hypotheses, but they still must return to experimental systems for validation.
The platform design also reflects the real bottleneck in AI drug development today. Anthropic said Claude Science comes with more than 60 curated skills and connectors, and includes review agents that can check citations and calculations; foreign media also reported that it can handle computational workflows, generate charts and drafts traceable to code and environments, and even present three-dimensional protein structures, chemical models, and genome browser tracks. In other words, the product’s selling point is not just the speed of answers, but whether the path behind those answers is sufficient to allow them to be rerun, reviewed, and questioned.
More sensitive is that Anthropic reportedly does not only want to serve pharmaceutical companies and research institutions, but also plans to enter drug discovery itself. The Verge quoted Eric Kauderer-Abrams, the company’s head of life sciences, as saying the goal will focus on exploring therapies for neglected diseases; Times of India also reported that the company will launch an internal preclinical drug discovery program. However, publicly available information remains limited, including the first disease targets, whether it will collaborate with laboratories, and how animal studies, clinical trials, and manufacturing will be arranged. Clear details have yet to emerge.
Background
The timing is unusual. Benzinga described the contrast in industry sentiment as “Anthropic Is Going All In on Drug Discovery Just as Pharma Stops Talking About AI”; on the same day, Anthropic also held an AI for Science event in San Francisco, bringing together pharmaceutical companies, research institutions, and company executives, and portraying Claude as infrastructure for compressing research timelines and building an AI-native scientific research environment. In market language, AI drug development is shifting from grand promises toward workflow integration; from a scientific perspective, that step instead demands greater transparency.
The limitations are therefore clearer as well. External experts have cautioned that AI has been used across multiple parts of drug discovery, but real-world experiments, toxicity assessment, clinical trials, and regulatory review remain the main gates; as of the publication of the related reports, no AI-designed drug had received approval from the U.S. FDA. If Claude Science can improve data analysis and decision records, it may make early-stage research more efficient; but whether candidate molecules are safe and effective will ultimately still have to be answered by reproducible experiments and human clinical evidence.