Anthropic describes ART system discovered with help from Claude agents
Anthropic says its Claude agents helped identify a previously uncharacterized ART system in DNA sequences. According to the company, preliminary laboratory results showed the expression of repeated DNA arrays into short RNA, but the system’s function remains unknown.

Anthropic announced the ART system on September 23, 2026, after its Claude agents identified it while analyzing DNA sequences. The company calls it array-associated reverse transcriptases, abbreviated as ART. According to the announcement, it is a previously uncharacterized biological system found primarily in bacteriophages. Anthropic has also established its own life sciences group and laboratory in the Bay Area.
The result is not yet a peer-reviewed scientific publication. The company released a technical report in preprint form and reported preliminary experimental observations. ART’s biological function has not been confirmed, and it is also unknown whether the system could be practically used, for example, in gene-related work.
What makes up the ART system
According to the technical description, ART consists of a reverse transcriptase, an adjacent partner gene, and a long array of regularly spaced DNA repeats. Reverse transcriptase is an enzyme that can create a DNA template from RNA. Anthropic says its agents identified this set of elements in the DNA sequences they analyzed.
According to the company, initial laboratory experiments showed that the DNA repeat array is expressed into short RNA. This is an early observation that does not yet clarify what the system does in bacteriophages or how its individual parts work together.
Anthropic hypothesizes that ART could function through a mechanism analogous to programmable systems that work with DNA. However, this interpretation remains a hypothesis. The published data do not yet confirm that ART is a gene-editing tool or that it has medical or biotechnological significance.
The role of Claude agents and the human laboratory
The company says the search took 21 hours and involved approximately 950 Claude agents. They used about 210 million tokens during the work. Their task was to search genomic data and identify patterns that warranted further investigation.
The physical experiments were not performed autonomously by the model. Human scientists conducted them in a laboratory and then tested selected hypotheses. The announcement is therefore an example of a workflow in which AI helps with large-scale data searches and the formulation of candidate explanations, while experimental confirmation remains the laboratory’s responsibility.
Anthropic does not present ART as equivalent to CRISPR or as a finished biological tool. The comparison with CRISPR concerns primarily the possible category of programmable systems, not confirmed capabilities of the new finding.
What still needs to be verified
Independent verification of ART’s existence, novelty, and function will be crucial in evaluating the result. Biochemical and structural experiments will be needed to clarify the system’s mechanism and the role of the short RNA produced from the repeat array.
A peer-reviewed publication, along with data or materials that allow other teams to reproduce the results, will also be important. Only subsequent experiments may show whether ART has programmable properties useful for work with DNA or RNA.
For now, the announcement documents the AI-assisted discovery of a candidate biological system and its initial laboratory testing. It is not, however, a confirmed new tool comparable to CRISPR or a clinical discovery.
Sources
- Anthropic – Confirms the laboratory’s announcement, the ART name, the description of the identified system, the scope of the agent search, human participation in the experiments, and the acknowledgment that ART’s function is not yet known.
- Anthropic technical report – The technical report is the primary source for the methodology and preliminary experimental results, but it does not constitute peer-reviewed verification.
- TechCrunch – Independently provides context that the physical laboratory work was not performed autonomously by Claude and that the significance of the discovery must be confirmed by the broader scientific community.
Verified and updated: September 24, 2026 06:20



