Anthropic has announced that it will operate its own drug development programs in the future – specifically for diseases that are economically unattractive for the traditional pharmaceutical and biotech industries. The announcement was made on June 30, 2026, at an event in San Francisco, where the company also introduced its new tool “Claude Science.”
Claude Science: a working environment for scientists
According to the official announcement from Anthropic, Claude Science is not a new AI model, but an application that consolidates common research tools and databases into a single working environment. A coordinating agent has access to more than 60 preconfigured tools and databases for areas such as genomics, proteomics, and cell biology and can bring in specialized sub-agents as needed. A separate review agent checks citations and calculations and marks or corrects errors. Each result is stored by Anthropic along with the underlying code and the complete conversation history, allowing analyses to be retraced and reproduced even months later. The beta version is available to users of the paid Claude plans (Pro, Max, Team, Enterprise) on macOS and Linux.
An example documented by Anthropic: Prasad Shirvalkar, associate professor of neurosurgery and anesthesiology at UCSF, reports that Claude Science immediately found a laboratory virus contamination in his team’s RNA sequencing data – a problem his team had struggled with for almost a year. According to several media outlets, including CNBC, Anthropic also presented an example at the event where Claude analyzed 100 rare genetic diseases in less than an hour and identified 32 candidates for computer-assisted screening.
Own drug research for neglected diseases
In parallel to Claude Science, Anthropic’s head of life sciences, Eric Kauderer-Abrams, announced according to CNBC that the company will launch its own early preclinical drug research programs. The focus is on diseases for which the traditional pharmaceutical and biotech industries show no interest for economic reasons, despite a real disease burden. Jonah Cool, Anthropic’s head of life sciences partnerships, characterized the initiative to CNBC as a complement to the core business with Claude Science: Only the company’s own experience in drug research allows for the development of the right models and tools for the entire industry.
Anthropic is organized as a Public Benefit Corporation, which, according to its own statement, gives the company more leeway to pursue research areas without clear commercial returns. Which specific diseases or drug targets are the focus, what budget is allocated for this, and whether Anthropic intends to bring drug candidates to market maturity itself or pass them on to partners has not yet been disclosed by the company.
How realistic are Novartis’ figures?
At the same event, Vas Narasimhan, CEO of Novartis and a member of the Anthropic board, also spoke. According to Narasimhan, the development of a drug from the finished drug candidate to approval currently takes about twelve years. He distinguishes between three types of delays: information latency, operational latency, and biological latency – the latter includes animal testing, cell models, and clinical trials in humans and can hardly be shortened by better software. The first two categories account for about 40 percent of the total time, according to Narasimhan; he sees the greatest potential for AI tools here, which could reduce the development time to seven to eight years. He estimates that the success rate could double from 8 to 16 percent, among other things due to better safety predictions.
These forecasts come from a Novartis manager who is also on the Anthropic board and are not yet independently verified study results, but rather his personal professional assessment. For context, Narasimhan also mentioned that large pharmaceutical companies invest a total of 150 to 200 billion US dollars annually in research and development and have brought about 800 to 1,000 drugs to market maturity over 120 years.
A competitive field: DeepMind, OpenAI, and Google are also in the game
Anthropic is not the only AI company investing in drug research. Google DeepMind CEO Demis Hassabis co-founded the company Isomorphic Labs with Alphabet, which specifically uses AI for the development of new drugs; DeepMind’s protein prediction tool AlphaFold has been regarded for years as a showcase example of AI in biology. Its co-developer John Jumper joined Anthropic in June 2026. OpenAI, on the other hand, has released its own model for biomedical research with GPT-Rosalind and introduced its own health section in the clinical field with ChatGPT Health.
Caution in clinical application
Regardless of the advances in drug research, experts continue to urge caution when AI is used directly in patient care – for example, in diagnoses or treatment decisions. Regarding two recently presented AI diagnostic systems that performed similarly well as doctors in simulations, Catherine Pope, a professor of medical sociology at the University of Oxford, expressed reservations: Real treatment situations are significantly more complex than controlled test scenarios can depict. Although this assessment does not directly refer to Claude Science, it points to a fundamental reservation regarding AI systems in immediate clinical use – an area that Anthropic’s new drug research programs do not address, as they are limited to the early preclinical phase.
Assessment
Anthropic’s foray into drug research can be read on two levels: as a contribution to the company’s nonprofit mission, but also as a strategic step to build credibility with those pharmaceutical companies that Anthropic wants to win as customers for Claude Science through practical experience in drug development. Whether the preclinical programs will actually yield market-ready drugs, whether Anthropic will further develop them itself or transfer them to partners, and whether Narasimhan’s forecasts regarding development time and success rate will be confirmed in practice remains to be seen.
