Anthropic Says Claude Found a New Enzyme System; CRISPR Researchers Call It Routine

Anthropic says its AI model Claude discovered a previously unknown enzyme system hidden in public DNA databases — and that the model performed most of the analysis on its own.

That claim, reported by The Decoder, immediately drew pushback from a very different framing: CRISPR researchers described the work as routine genome mining, according to the report's headline framing of the dispute. The gap between those two characterizations is the story. It is a fight over language that the AI industry uses for scientific credit, and it arrives at a moment when AI labs are aggressively marketing their models as engines of autonomous scientific discovery rather than as tools that assist human scientists.

What Anthropic says happened

According to the report, Claude identified an enzyme system that was previously unknown, working from DNA sequence data held in public databases. Anthropic's claim about the work has two components, and both matter. The first is the scientific claim: a novel enzyme system exists in the data, and nobody had catalogued it before. The second is the methodological claim: Claude did most of the analysis itself, with limited human direction.

The second claim is the more contentious one. If an AI model can autonomously survey genomic databases, detect a pattern that human researchers missed, and characterize a functional system from sequence data alone, that changes the economics of biological research. Genome databases have grown faster than the human expertise available to interpret them, and any system that can comb through that backlog without a principal investigator directing every step would have real research value. Enzyme discovery feeds directly into biotechnology applications, including protein engineering, industrial biocatalysis, and the broader field of genome editing where CRISPR itself was born.

The skeptics' case

The counterargument from CRISPR researchers, as captured in the report, is blunt: this is genome mining, a practice molecular biologists have performed for decades. Genome mining — searching sequence databases for novel genes, enzyme families, and biosynthetic pathways — is an established discipline with its own conferences, software tools, and career tracks. Researchers routinely discover new enzymes and enzyme systems by running computational searches against databases such as those holding microbial DNA. From this perspective, an AI model doing the same thing is not discovering in any new sense. It is automating a known workflow.

The dispute therefore turns on two questions the report leaves open. First, is the enzyme system genuinely novel in a scientifically meaningful sense, or merely unannotated — the kind of finding that falls out of a competent database search? Second, does the degree of autonomy matter? If Claude directed the analysis end to end and the human role reduced to framing the question and verifying outputs, that is a different kind of result than a human researcher using an AI as a faster search engine, even if the underlying finding is identical.

Why the wording matters

For Anthropic, the framing of a model "discovering" something is commercially and politically consequential. AI labs are competing to demonstrate that their models can produce genuine scientific output, not just fluent text. Claims of AI-driven discovery support arguments about the technology's economic value, inform ongoing policy debates about AI's impact on research and labor, and shape how laboratories decide whether to integrate these tools into their pipelines.

For working scientists, loose use of the word "discovery" carries a different cost. Scientific credit is a tightly governed currency. A claimed discovery normally requires peer review, reproducible methods, and validation of the biological function — not just a computational annotation of a database hit. Enzyme systems in particular require experimental confirmation that the predicted proteins do what the analysis suggests. The researchers' pushback reflects a concern that AI marketing language can inflate computational findings into discoveries before the wet-lab work exists to support them.

The disagreement also illustrates a structural feature of the current moment in AI and science. Models like Claude are powerful at pattern detection across large corpora, including biological sequence data. Pattern detection is a genuine component of discovery. But the scientific community distinguishes between finding a candidate in a database and demonstrating that the candidate does something. The CRISPR researchers' characterization of the work as routine genome mining is, in effect, an argument about where on that spectrum the finding sits.

What to watch

The Decoder's report does not indicate whether the claimed enzyme system has been experimentally validated or submitted for peer review, and those two steps will determine which framing survives. If the finding clears peer review and the enzyme system proves biologically real, the autonomy claim becomes the interesting part and Anthropic's position strengthens. If the work remains a computational result, the CRISPR researchers' description — competent mining of public data — will likely stand as the more accurate account.