Reported by 1 source

The short version

  • Anthropic’s Claude autonomously identified a novel enzyme system in bacteriophages that shares structural similarities with early CRISPR components.
  • While the discovery highlights advances in A.I. autonomy, researchers caution that the biological function and potential biotechnological applications of the new system remain unknown.
  • The finding has sparked debate regarding data privacy and whether user-shared research content may have inadvertently influenced the model's discoveries.

Anthropic has announced that its artificial intelligence system, Claude, autonomously discovered a novel enzyme system within bacteriophages. The discovery, detailed in a preprint study released in late September, describes an array-associated reverse transcriptase, or ART, which shares structural characteristics with the components of CRISPR. This finding represents a notable advancement in the capacity of A.I. agents to conduct independent biological exploration, although scientists emphasize that the practical significance of the new system remains unproven.

The research process involved instructing approximately 950 A.I. agents to search through extensive DNA databases for reverse transcriptases. These enzymes are responsible for reading RNA strands and creating complementary DNA sequences. The agents were programmed to write their own code to facilitate this search, initially identifying roughly 200,000 such enzymes. After operating for about 21 hours, one agent flagged an unusual reverse transcriptase accompanied by a partner gene and a long array of repeating DNA sequences.

News Journal

This specific configuration mirrors the foundational elements of CRISPR technology. CRISPR was originally identified in 1987 within E. coli bacteria due to its repeating DNA sequences, which help microbes defend against viral infections. However, it took decades for researchers to understand how to harness this system for precise gene editing. The current A.I. discovery is more comparable to those initial observational findings from forty years ago rather than the mature, revolutionary gene-editing tool used today.

Despite the structural similarities, Anthropic scientists have not determined what the newly found enzyme system actually does. They cannot confirm whether it possesses any utility comparable to modern CRISPR applications, such as developing personalized medicines for genetic disorders. Dario Amodei, CEO of Anthropic, acknowledged this uncertainty in a social media post, stating that the precise function and biotechnological value of the system are not yet clear. He noted, however, that he would have been proud to achieve such a discovery during his own doctoral studies.

Independent experts have praised the methodological breakthrough even if the biological implications are still emerging. Feng Zhang, a molecular biologist at the Broad Institute of MIT and Harvard who was not involved in the study, described the work as an exciting example of how A.I. agents can contribute to biological discovery. He expressed hope that this effort would encourage more scientists to explore how artificial intelligence can support their research endeavors.

Stanley Qi, a CRISPR researcher at Stanford University, highlighted the system's ability to recognize complex biological patterns that were previously difficult to detect. He noted that nature contains a vast diversity of molecular systems that are not yet understood, and A.I. could significantly expand the speed and effectiveness with which scientists explore these areas. Dimitri Perrin, a computer scientist at Queensland University of Technology, also characterized the work as a major advancement in A.I. autonomy.

The announcement has raised questions regarding data privacy and potential conflicts of interest. Mario Rodríguez Mestre, a computational biologist at the University of Copenhagen, reported that his team has used Anthropic’s A.I. tools for three years to write code and draft manuscripts while researching ARTs for four years. This overlap prompted concerns about whether Claude might have benefited from content shared by users during their interactions with the platform.

Anthropic addressed these concerns in a statement, asserting that they are not aware of any previously published work describing the specific ART system found by Claude. The company clarified that the model was not trained on user transcripts and that its molecular biology team does not have access to such data. This controversy follows similar debates in mathematics, where A.I.-assisted discoveries led to disputes over credit and prior knowledge among researchers.

The broader context of this discovery includes recent developments in other fields where A.I. has played a role in solving longstanding theoretical problems. For instance, earlier this year, an Anthropic employee used Claude to help disprove the Jacobian conjecture, a decades-old mathematical problem. These events collectively underscore the growing integration of artificial intelligence into fundamental scientific research, even as questions about attribution and data integrity continue to emerge.

As the scientific community reviews the preprint study, the focus remains on validating the A.I.'s findings through traditional laboratory methods. While the initial excitement centers on the autonomy of the discovery process, the ultimate test will be whether this new enzyme system offers any tangible benefits for biological research or medical applications. Until then, the discovery stands as a proof of concept for how machine learning agents can navigate complex biological data.

Sources behind this briefing

Go to the original reporting

  • Smithsonian Magazine↗Anthropic Says Its A.I. Discovered a New Enzyme System That Resembles the Revolutionary Gene-Editing Tool CRISPR