Anthropic says Claude found a phage enzyme system whose job is unknown

Why it matters

Anthropic is moving Claude's biology work from literature and data analysis toward a lab that can test hypotheses. The phage finding is an early lead, while its unknown function makes clear that a model-generated candidate is not yet a validated discovery.

Anthropic says Claude identified a previously unknown enzyme system in bacteriophage DNA, with a nearby array of repeating DNA that the company says resembles CRISPR. Anthropic described the finding as the first result from its new molecular biology lab in a two-post thread on X on September 23rd. The system's function is still unknown, the company said.

That qualification sets the limits of the announcement. Anthropic has not described a function for the enzyme system or reported a biological experiment establishing what it does. The post says a team of Anthropic biologists uses Claude to explore and accelerate fundamental biology research. Claude works through data and scientific literature to generate hypotheses and identify candidate biological systems, according to the company. The thread does not name the enzyme, specify the dataset, or explain how the candidate was checked against existing research.

The result arrives days after reporting established that Anthropic has a wet lab in the San Francisco Bay Area, where researchers can conduct physical biology experiments. In a September 18th Reuters report, Anthropic's head of life sciences, Eric Kauderer-Abrams, said the company does some work in its own facilities and some with external partners. He described AI's role in executing lab work as being in its early stages. Anthropic has not said in this thread whether the phage finding has been tested at the bench.

That distinction gives the announcement its real weight. A model that finds a pattern in genomic data has produced a research lead; demonstrating what the biological machinery does requires further work. Anthropic itself says the enzyme system's purpose is not understood. The company's post presents the finding as a candidate for investigation, not as a validated mechanism or a practical technology. Similarity to CRISPR describes the nearby repeating DNA structure, as Anthropic put it; it does not establish that the system performs CRISPR's functions.

For Anthropic, the lab gives Claude's biology work a path from digital analysis toward experiments. Kauderer-Abrams told Reuters that real lab work remains the final test in biology and said Anthropic is building in-house capacity partly to move faster and gain direct experience. The bacteriophage result is an early example of the hypothesis-generation step the company wants Claude to support. Whether the lab can turn such candidates into reproducible findings is a separate test.

The effort also fits a personal rationale for Anthropic co-founder and CEO Dario Amodei. Reuters reported that Amodei has linked his interest in life-sciences research to his father's death from a disease before a cure became available. The company has said it wants to apply AI to treatments for conditions that receive less attention from pharmaceutical developers. A phage enzyme, whose function remains unknown, is far from evidence of a medical application; its immediate value to Anthropic is showing the kind of biological lead the lab and Claude might pursue.

There is a second pressure in Anthropic's move into biology: the same capabilities that can assist legitimate research raise questions about misuse. In its own account of biological risk, Anthropic has described biology as dual-use and said it applies safeguards to certain high-risk biological requests. That makes the company's claim of discovery only one part of the story. The next evidence to watch is whether researchers can identify the system's role, test it experimentally, and explain how Claude contributed beyond searching data and proposing a hypothesis.