Anthropic's AI Discovery and Its Implications for Biomedical Research

<p>Dario Amodei, the chief executive of Anthropic, stated that AI has the potential to cure most diseases within the next decade. In late September 2026, he announced that the company's AI model, Claude, discovered a new enzyme system within 24 hours, which the company claims resembles CRISPR, a gene-editing tool currently used in medical treatments. During the announcement, Anthropic suggested that this discovery could revolutionize medicine and quoted a gene-editing pioneer who described it as 'genuinely intriguing.'</p><p>However, the company acknowledged uncertainty regarding the significance of the finding, which originated from a biology lab established by Anthropic earlier in the year. Some academics, including the CEO of Eli Lilly, criticized the portrayal of the finding as a major breakthrough, with Fyodor Urnov, a scientist involved in a CRISPR project with Anthropic, stating that 'the jury is out' on whether the discovery is indeed an enzyme. Multiple researchers informed <em>The New York Times</em> that they had been aware of the enzyme system for years, including a team at the University of Copenhagen, which alleged that Anthropic's results might have been derived from their prior interactions with Claude. Anthropic responded that Claude is not trained using user transcripts and did not comment further on the matter.</p><p>Despite the grand claims, the actual achievement of Claude may not lead to significant biomedical advancements. Nevertheless, Claude's ability to analyze vast amounts of biological data is noteworthy, particularly for scientists overwhelmed by the volume of data generated in recent years. This capability, while not miraculous, is valuable for researchers facing challenges in data analysis.</p><p>At the turn of the 21st century, the Human Genome Project sequenced all 3 billion letters of human DNA, taking 13 years and costing $2.7 billion. Today, sequencing a full genome takes just over a day and costs around $200. Consequently, the amount of publicly available genomic data has grown to unmanageable levels. In other fields, such as protein science, similar data collection booms have occurred.</p><p>Extracting useful insights from this data has become a long-term challenge. A 2012 committee of bioinformatics experts noted that the 'bottleneck in scientific productivity' had shifted from data production to data analysis, but early initiatives at the National Institutes of Health to address this issue did not succeed. Efficiently storing, transmitting, and examining the generated information could enable researchers to develop new drugs by identifying better genetic targets. However, the sheer volume of data makes comprehensive analysis difficult.</p><p>AI agents are well-suited to handle large biological datasets. Bo Wang, an assistant professor at the University of Toronto and chief AI scientist at a drug-discovery company, stated that AI's significant impact in biology will come from its ability to 'scale scientific attention.' By deploying AI on extensive datasets, researchers can identify outliers that merit further investigation. AI can also propose and test hypotheses regarding unusual patterns and assist scientists in selecting experiments to conduct in the lab. Automated labs, some of which already exist in the U.S., may eventually perform these experiments, with substantial funding allocated for their establishment.</p><p>AI is already influencing medicine, with models that analyze patient medical histories to predict health risks, including cancer susceptibility, being utilized in most U.S. hospitals. However, not all biological fields generate sufficient data for AI applications. For instance, studies on cellular communication may not benefit as much from AI due to limited microscopy data availability, according to Stirling Churchman, a genetics professor at Harvard Medical School. Other areas, such as organelle biology, require extensive and time-consuming experimentation.</p><p>Churchman noted that 'there will be magic where there’s data.' She spent the summer at OpenAI developing a specialized model for life-science researchers, anticipating that AI will enable non-coding scientists to analyze data significantly faster, allowing more focus on biological inquiries and experiments.</p><p>The recent finding by Anthropic illustrates how AI can uncover previously unnoticed outliers. Existing tools can analyze large genomic datasets and have already identified new gene-editing systems. AI's ability to conduct these analyses at a larger scale, while evaluating intermediate results, was emphasized by Kendell Clement, who leads a computational-biology lab at the University of Utah. However, achieving this scale requires substantial resources, as demonstrated by Anthropic's deployment of approximately 950 Claude bots to analyze a database with nearly 2 billion DNA entries.</p><p>AI companies are motivated to promote the medical breakthroughs their models may facilitate, especially amid growing skepticism. Amodei suggested that curing cancer would help alleviate concerns about AI's capabilities. While such a breakthrough would be significant, enhancing scientists' efficiency is also valuable, even if it lacks the same level of excitement.</p>