Anthropic says its Claude AI model discovered a new CRISPR-like gene editing system, though scientists still don't know exactly what it does

[Nina26]

Anthropic announced on September 23, 2026 that its Claude AI model helped discover a previously uncharacterised biological system called array associated reverse transcriptases, or ART, found primarily inside bacteriophages, the viruses that specifically infect bacteria. The system consists of three distinct parts, a reverse transcriptase enzyme, an associated partner gene, and an array of evenly spaced DNA repeats whose layout closely resembles the repeat structures found in CRISPR systems, the gene editing technology that has already reshaped modern biology.

The discovery came from a research effort in which 950 separate Claude agents spent roughly 21 hours collectively using 210 million tokens to search through massive DNA sequence databases, gathering over 200,000 reverse transcriptase enzymes, identifying 3,500 new candidate systems, and eventually narrowing that pool down to the 20 most compelling candidates for detailed human review, a process Anthropic says would typically take human scientists weeks or months to complete manually. Human researchers then validated the most promising findings and conducted laboratory testing, with early experiments showing the ART array expresses distinct short RNA molecules, suggesting the system may have programmable functionality similar to CRISPR itself.

CRISPR pioneer Feng Zhang, commenting on the discovery, called it an exciting example of how AI agents can contribute to biological discovery, adding that the findings merit further investigation. Anthropic was notably candid about the discovery's current limitations, acknowledging that the system's primary biological function remains genuinely unknown, and that researchers still have ongoing work ahead of them to understand exactly how ART operates at a mechanistic level before its real world applications, if any, can be properly determined.

AI model Claude discovers CRISPR-like enzyme system, Anthropic sayswww.aljazeera.com

[BretHart_99]

950 separate AI agents collectively completing in 21 hours what would normally take human researchers weeks or months is a genuinely staggering illustration of how much AI can accelerate the earliest, most labour intensive stages of biological discovery specifically.

[Scholar]

Anthropic being this upfront that the system's actual function remains unknown is refreshingly honest, it would have been easy to oversell this as a finished breakthrough rather than clearly framing it as an interesting early lead that still needs a lot more research.

[Ori10]

Feng Zhang's endorsement carries real weight given his own history with CRISPR itself, having someone that credentialed call the findings worth further investigation lends this considerably more scientific credibility than a purely corporate announcement would have on its own.

[Voyager17]

The repeat array resembling CRISPR's own structure so closely is the detail that makes this genuinely exciting rather than just another obscure enzyme discovery, that specific similarity is exactly what raises the possibility of real programmable gene editing potential down the line.

[Oliver87]

AI narrowing 200,000 candidate enzymes down to just 20 worth serious human attention is the actual value proposition here, that kind of dramatic filtering of an enormous search space is precisely the type of task AI seems genuinely well suited to accelerating.