Earlier this week, at an event called "The Briefing: AI for Science," artificial intelligence company Anthropic announced the launch of its new Claude Science platform, positioning it as an "AI workbench" for scientists. The platform is designed to integrate disparate scientific research tools and data sets into the same environment and automatically generate charts and visualization results. Anthropic believes that this type of AI technology has the potential to "significantly accelerate the development of scientific discoveries and medical interventions." The company also emphasized that a large number of biotechnology and pharmaceutical companies are already using Claude for related research and development work.
Even more ambitiously, Anthropic is no longer content to be just a tool provider and has publicly stated that it plans to develop drugs itself. Eric Kauderer-Abrams, head of life sciences, said in an interview that the company will focus on “neglected” disease areas and hopes to discover new treatment options with the help of AI. In the current boom in AI drug research and development, technology giants such as OpenAI, Amazon, and Google have all launched their own life science platforms. However, Anthropic’s statement is one of the few public moves by a cutting-edge general AI model company to directly announce that it will develop its own drugs. This also puts it in a rather delicate position: on the one hand, it sells software tools to many pharmaceutical companies, and on the other hand, it may become a potential competitor of these customers in drug research and development.
Industry insiders pointed out that Anthropic’s move actually pushed the company into a broader competition. Joining the competition are “AI-first” drug companies such as Insilico, Isomorphic Labs, a spin-off from Google DeepMind, and a large number of traditional biotech and large pharmaceutical companies that build or acquire AI tools. Despite its huge momentum, Anthropic has so far provided very limited specific information: they have not stated how they will advance if they find promising drug candidates, nor have they responded to detailed questions about which diseases will be targeted in the first batch, or whether they will cooperate with other institutions to complete laboratory research, animal testing, clinical trials, and manufacturing.
Behind the words “AI drug discovery” is an extremely broadly defined concept. Namshik Han, a professor at the University of Cambridge and co-founder of the AI biotechnology startup CardiaTec, believes that AI has been used in "every stage" of drug discovery, from screening and optimizing new compounds to assisting scientific research, data analysis, clinical trials and even production and manufacturing. Matthew Todd, professor of drug discovery at University College London, also expressed a similar view, saying that "AI drug discovery" has almost become a "catch-all term" to refer to this entire set of broad application scenarios.
Still, experts generally agree that AI is still in its early stages of changing drug development. Han pointed out that pharmaceutical giants such as AstraZeneca, Novo Nordisk, and GSK have deployed many AI projects, using models to generate possible candidate molecules for known pathological pathways or existing targets, helping researchers discover new molecular structures that can interact with specific receptors. Todd emphasized that AI is very useful in accelerating scientific research and helping "road test" new drug ideas. It can screen potential solutions in the huge chemical and biological space and find connections that are difficult or take a lot of time to discover under human and traditional tools. Combined with Anthropic's advantages in cutting-edge models, it is generally speculated that it will mainly use generative AI to search and recommend among massive combinations of compounds and biological targets, assisting researchers in proposing new drug design ideas, identifying new disease targets, or finding new indications for existing drugs.
However, there is still a long distance between “proposing a drug idea” and “actually having a drug enter clinical practice and be approved for marketing.” Todd said that it is still "far away" from the first drug designed entirely by AI and passed regulatory approval to enter the market. The entire discovery and development process will not be fully automated for a long time, and the participation and supervision of human experts will still be necessary. Todd and Han also pointed out that the lack of a large amount of public, high-quality experimental data - especially detailed records of the specific behavior of compounds in the human body - constitutes a key bottleneck at this stage. Even in the most deeply studied fields of biology, there are still huge gaps in human understanding of many mechanisms.
Frank von Delft, professor of structural chemical biology at the University of Oxford and head of protein crystallography at the Oxford Center for Drug Discovery, believes that the public's expectations for powerful AI models are justified, but current technology is "nowhere near the point where experiments are no longer necessary." Drug candidate molecules still need to be tested in the real world regarding efficacy, toxicity, formulation, storage and safe administration. These links require a large number of professionals to invest huge amounts of money and time. Especially when conducting human clinical trials, many seemingly promising drug candidates often fail. von Delft bluntly stated that if Anthropic really wanted to develop the drug itself, "it would have to invest heavily in experiments."
Judging from recent trends, Anthropic seems to be preparing for this. In the past year, the company has continued to recruit biology-related talents and is preparing to build its own wet laboratory. At the same time, it has released multiple life science job recruitment information on the public recruitment platform. Han revealed that Anthropic “has been actively recruiting” in this area and has even extended offers to several academic colleagues. He said he understands that Anthropic has successfully poached some professionals from large pharmaceutical companies and some top academic institutions, but did not disclose specific names.
Even so, with such a complex R&D system, no matter which disease Anthropic ultimately chooses to target, it will be many years before the results are truly seen. Taking the traditional drug development cycle as an example, it often takes close to ten years for a new drug to complete clinical trials. Todd said that "there is always a huge time lag" in drug testing because it is a time-consuming and lengthy task to prove the safety and effectiveness of a certain drug candidate through experiments. So far, no drug designed by AI has successfully completed all clinical trials and received FDA approval for marketing. A few candidate drugs have entered the clinical stage, but it is difficult for the outside world to accurately judge the specific role that AI has played in its development process, and whether these drugs can significantly surpass products with traditional R&D paths in efficacy.
According to experts, what AI can significantly speed up is the "search" and "imagination" parts, but what really determines the success or failure of a drug is still those experiments and tests carried out in a rigorous and slow manner in the real world. For Anthropic, this means a huge role change from imagining AI rewriting drug discovery to taking on high-investment, high-risk, long-term experimental responsibilities.