“[The DNA next to the RT] is spectacular: I can see by eye a tandem repeat array … that's a CRISPR-like … repeat array?!”
That is what one of the agents wrote while reading the raw DNA next to an odd-looking gene. Anthropic quotes it in its announcement and does not hide it, because at that moment the machine does what a scientist usually does: it notices a pattern that should not be there.
Fact
The facts: on 23 September 2026 Anthropic introduced its fundamental biology research group and lab, and its first result. About 950 Claude agents searched a database of DNA sequences for 21 hours using 210 million tokens; they gathered over 200,000 reverse transcriptases (enzymes that copy RNA into DNA), picked out 3,500 new candidate systems and wrote up the 20 most promising in reports for the scientists. One of the agents spotted an array of repeating non-coding DNA sequences next to an unusual reverse transcriptase from a so-called jumbo phage. The enzyme itself was known from earlier studies, but per Anthropic, Claude is the first to notice the associated array and the additional protein of unknown function. The company calls the system ART (array-associated reverse transcriptases) and reports that first experiments show the array is expressed as a set of distinct short RNAs. Its function is not yet known; a preprint has been published. Feng Zhang of MIT and the Broad Institute, one of the pioneers of CRISPR genome editing, calls the finding “genuinely intriguing”. The lab work is done by people, in a lab at biosafety levels BSL-1 and BSL-2.
What the agent saw
The speed is expected, however impressive: twenty-one hours against weeks and months of human work. The more interesting part is that the agent chose on its own to look closer at something nobody had pointed it to, checked whether it had been described, and wrote a report for human review.
And just as important is what did not happen. The system's function is unknown. It is not a drug, not a tool, not a new CRISPR. For now it is a strange pattern in the DNA of a virus that infects bacteria, and first experiments showing the array is expressed as distinct short RNAs.
The discovery is a question, not an answer. And this time the machine asked it.
The answer will come from the peer-reviewed version and from other people's labs. If ART turns out to be programmable like CRISPR, this date will be remembered. If not, the proof remains that a machine can notice.
The visual is generated code-art, no third-party images.