Anthropic’s AI Biology Breakthrough Sparks Questions Over Research Origins
Last week, Anthropic, a prominent player in the artificial intelligence sector, presented findings from its newly established biology lab. The scientists reported that they utilised AI agents to identify novel enzymes that hold potential for advancements in biotechnology. Some scientists were swift to express scepticism regarding the accomplishment. Mario Rodríguez Mestre, a computational biologist, stated this weekend that he and his colleagues have been examining the enzymes and their associated molecules – referred to by Anthropic as ARTs – for a duration of four years. Dr. Mestre and his colleagues have not yet disseminated their findings. For the past three years, they have consistently utilised Anthropic’s AI models in various capacities, including coding, manuscript drafting, and other tasks. In this context, Dr. Mestre indicated that he and his colleagues communicated significant discoveries regarding the enzymes to Anthropic. “So, for me, the most important question is not ‘Were ARTs already known?’ They were,” Dr. Mestre said in an interview. The issue instead, he added, is whether Anthropic’s AI actually reasoned its way to the results, or whether it was guided to them based partly on his own work. “My concern is that this information was used to train future versions of the models,” Dr. Mestre said of his own research. “I think it’s important to raise this possibility.” In a statement on Sunday, Anthropic said: “We are not aware of any previously published work describing the ART system we recently found. Claude was also not trained on any user transcripts, and our molecular biology team has no such access, either.”
Anthropic stated that Claude’s significant contribution was the identification of a variety of RNA molecules and a protein linked to reverse transcriptase within a system. Dr. Maestre noted that he and his colleagues had made that discovery over a year ago. Concerns have emerged once again as researchers at OpenAI announced a resolution to a long-standing mathematical conundrum. A prominent mathematician observed that he had been utilising the company’s AI in pursuit of a solution for an extended period. Anthropic, anticipated to enter the public market with a valuation that could reach $2 trillion, has emphasised biological research as a fundamental aspect of its mission. “We believe this type of application will accelerate rapidly in the years ahead, and that we can make profound progress in the near future on diseases that have plagued humanity for millennia,” Dario Amodei said in a speech to the United Nations on Wednesday. In a report published last week, researchers from Anthropic indicated that they directed AI agents to explore databases housing billions of gene sequences. The scientists refrained from detailing the specific prompts employed in their research. The databases encompass DNA gathered from marine environments, terrestrial soil, and various other ecosystems. The genes are predominantly associated with microbes and viruses; researchers have only ascertained the functions of a small subset of them.
Anthropic’s scientists assigned the AI the objective of identifying genes responsible for the production of enzymes known as reverse transcriptases. These enzymes interpret the sequence of an RNA molecule and synthesise a corresponding DNA molecule with an identical sequence. Bacteria utilise certain reverse transcriptases as a mechanism to detect invading viruses. Some viruses utilise reverse transcriptases as mechanisms to combat other viruses. The scientists at Anthropic supplied the AI with five instances of reverse transcriptases. The agents developed proprietary software to investigate the databases for analogous genes and to scrutinise potential candidates in detail. The agents identified genes for potential reverse transcriptases in viruses that infect bacteria, referred to as jumbo phages. Each of those sequences was situated in proximity to a cluster of genes responsible for RNA molecules, which could facilitate the enzymes in executing their functions. The Anthropic scientists named the enzymes array-associated reverse transcriptases, or ARTs. In a report posted to their website, they said that their AI had taken “the first steps of a biological discovery on its own.” On Sunday, Dr. Mestre disputed that claim. “They are the same systems we have been studying for years,” he said. And they are the same enzymes, he said, that he had described to Anthropic’s AI.
Dr. Mestre first encountered the enzymes in jumbo phages in 2022 during a quest for novel reverse transcriptases. He worked briefly as a consultant for ReNegade Therapeutics, which filed a patent in 2023 on several reverse transcriptases, including those identified by Dr. Mestre. He is identified as a co-inventor on the patent filing. After his tenure at ReNegade, Dr. Mestre transitioned to the University of Copenhagen to conduct an in-depth examination of reverse transcriptases, particularly those identified in jumbo phages. He and his colleagues referred to them as “jumbotrons.” Anthropic released Claude in 2023, a year subsequent to Dr. Mestre initiating his research on jumbotrons. Similar to numerous scientists, he deemed it beneficial for a variety of tasks. “I completely switched my way of doing science,” he said. In discussions with Claude, Dr. Mestre presented a draft of his dissertation alongside a manuscript detailing jumbotrons. He and his colleagues uploaded additional details while utilising Claude to execute various administrative tasks. Source examined Slack messages, figures, and various materials generated by Dr. Mestre that detail his research and discussions with Claude over the preceding years. They appear to indicate that the AI played a significant role in his research.
Seth Shipman, a bioengineer at the Gladstone Institutes and the University of California, San Francisco, acknowledged his familiarity with Dr. Mestre’s work. “We’ve been talking with him about these for years,” he said. “Many of us know about these systems.” Dr. Mestre expressed uncertainty regarding whether Anthropic had utilised his research to pursue jumbotrons. However, he expressed scepticism regarding the ability of the AI agents to locate them with such speed. Jumbotrons are uncommon, and their RNA genes present challenges in differentiating them from arbitrary genetic sequences. Dr. Mestre also identified notable parallels between his research and the findings presented by the Anthropic scientists. The scientists asserted that their AI had autonomously acquired the raw data from a 2022 experiment in which researchers documented all the viral RNA molecules generated by a microbe infected by jumbo phages. Anthropic’s AI indicated that the microbes produced elevated concentrations of RNA molecules in ARTs. The scientists contend that this evidence bolsters their hypothesis that the reverse transcriptases they assert to have identified depend on these RNA molecules to perform a function for the jumbo phages. Dr. Mestre indicated that he and his colleagues conducted an analysis of the same data a year prior, which facilitated their identification of the RNA genes linked to the jumbotrons. They disclosed this non-public information to Claude. “The resemblance was striking to us,” Dr. Mestre said.
Without a detailed examination of the actions taken by Anthropic AI, Dr. Mestre stated that it would be impossible to definitively conclude that it was influenced by his research. “I mean, coincidences happen,” he said. “Everyone in the field is doing more or less the same things. But we were aware of these repeats more than one year ago.” Dr. Mestre noted that AI companies also have been accused of training models on the work of artists and writers without copyright permission. “It’s already happened,” he said. Dr. Shipman, of U.C.S.F., said that he and his colleagues, too, have leaned heavily on Anthropic’s tools for their research. “We’re also telling them what we’re working on,” he said. The company’s announcement last week raised significant concerns among scientists who study these enzymes, he added. “I’ve heard from a bunch of people in our field who say, ‘Uh oh, are they training on my prompts?’” Dr. Shipman said. “We can’t not think about it.” The jumbotron controversy mirrors a previous scientific dispute that unfolded earlier this month. Another artificial intelligence start-up, OpenAI, asserted that it has solved a significant, previously unproven problem in mathematics. However, the mathematician from New York University, Tristan Buckmaster, who had been collaborating with a colleague on his own solution, indicated that they had utilised OpenAI’s models in their research.
In a statement, Dr. Buckmaster raised the possibility that OpenAI had borrowed from their efforts. “Is it ethical to use customer’s data to try to scoop their customer?” he later asked on social media. OpenAI subsequently refuted the assertion that its models had been trained on Dr. Buckmaster’s conversations. The New York Times has initiated legal action against OpenAI and Microsoft, alleging copyright infringement concerning news content associated with AI systems. The two companies have refuted those assertions. Dr. Mestre had developed a growing enthusiasm for the application of AI in his scientific endeavours. Just last month, he commenced utilising Claude Science, a novel AI tool developed by Anthropic specifically designed for researchers. However, following Anthropic’s announcement last week, he is scaling back his projects with Claude and transitioning to alternative models. “I’m shutting down everything,” he said.
