Anthropic Claude AI Discovery Sparks Debate Over Unpublished Research and Scientific Ethics

Anthropic Claude Claims Breakthrough in Biological Discovery

AI company Anthropic recently announced that its Claude system had identified an unusual family of enzymes called array-associated reverse transcriptases (ARTs) in bacteriophages—viruses that infect bacteria1

1

. The finding was presented as an example of AI initiating biological research, showcasing the potential for autonomous discovery in scientific research. However, the announcement quickly sparked controversy when a researcher raised concerns about whether unpublished work shared with Claude may have influenced the AI-generated findings.

Researcher Questions AI's Independent Discovery Claims

Mario Rodríguez Mestre, a PhD candidate in computational biology at the University of Copenhagen, revealed that his team has been investigating these same enzymes—which they call "jumbotrons"—since 20222

2

. According to The Scientist magazine, Mestre shared his dissertation and manuscript drafts with Claude during his research process. This disclosure raises critical questions about whether Anthropic Claude independently recognized the significance of these enzymes or whether unpublished work shared with Claude influenced its results. Mestre is named among the inventors of a patent for engineered "retrons"—genetic systems containing a reverse transcriptase enzyme and associated RNA—with a US application published in July 20231

1

.

What Anthropic Actually Discovered

The enzyme at the center of the AI discovery had already been identified by scientists from Texas A&M University2

2

. What Anthropic claims Claude recognized is a larger biological system: the enzyme, a series of repeated DNA sequences, and a possible partner protein. The researchers found these repeated sequences produce small RNA molecules, with some evidence coming from reanalyzing data published by another research team in 20221

1

. Human scientists conducted the laboratory experiments, and important questions remain unanswered. Researchers have not yet demonstrated that this enzyme is active, that it copies the associated RNA molecules into DNA, or what the system does in bacteriophages. Its resemblance to CRISPR, the well-known gene-editing tool, has attracted attention, though this does not confirm ARTs can edit genes.

Anthropic Denies Access to User Transcripts

Anthropic firmly denies that Claude was trained on user transcripts or that its biology team had access to them2

2

. The company maintains separate policies for personal and business services, with commercial services excluding inputs and outputs from training by default, though exceptions may apply when users give permission or submit feedback. However, there is no established evidence that Mestre's unpublished material was used in the discovery. The available reporting cannot independently resolve whether data sharing occurred inappropriately, leaving the scientific community to grapple with questions about transparency in AI-assisted research.

Verification Process Reveals Limitations in Autonomous Discovery

The Anthropic report about Claude's findings reveals significant limitations in claims of autonomous discovery. In 10 additional search runs, the AI agents missed the repeat array entirely1

1

. Some models could recognize the pattern when given relevant sequences directly, but recognizing a supplied pattern differs fundamentally from finding it independently. This inconsistency limits claims of a consistently autonomous discovery process by Claude, though it does not invalidate the biological finding itself. The Anthropic report has not yet been peer reviewed, raising further questions about the verification process for AI-generated findings.

Ethical Guidelines for Using AI in Research Remain Critical

Before sharing unpublished research with an AI tool, scientists should consult their university's guidance and seek training on responsible use2

2

. Guidelines at the University of Manchester advise researchers to protect confidential information, verify sources, and keep records of their AI use. A paid subscription does not automatically ensure that scientific research remains private or is excluded from model training. Researchers should record the model and instructions used whenever AI is essential to the method, developing documented analysis that other scientists can check and repeat. The controversy underscores why ethical guidelines for using AI must evolve alongside the technology, protecting both intellectual property and scientific integrity in an era of AI initiating biological research.