Global AI, Malaysia Angle / AI news for Malaysia
Anthropic says Claude agents found a new CRISPR-like enzyme system on their own
Anthropic says hundreds of Claude agents combed a huge DNA database and spotted an enzyme system nobody had described. The lab work is still early, and its function is unknown.

In brief
- Anthropic said on 23 September 2026 that Claude agents, given one prompt, found a previously undescribed enzyme system in phage DNA and filed it for human review.[1]
- The preprint says the agents surveyed reverse transcriptase genes across 1.9 billion protein clusters. The new family, called ART, carries CRISPR-like repeat arrays.[2]
- No Malaysian group is involved. Local efforts such as MyGenom and the Malaysia Genome and Vaccine Institute hold the kind of sequence data such agents are built to search.[5][4]
What Anthropic reported
Anthropic has set up its own biology laboratory and says its first research campaign has already turned up something new. Claude agents searching a large database of DNA sequences flagged an unfamiliar enzyme system that sits next to a long run of repeated DNA, a layout that resembles the arrays that make CRISPR gene editing programmable.[1][2]
The company is careful about what it is claiming. Its scientists do not yet know what the system does, and the findings are in a preprint that has not been peer reviewed. The claim being tested is narrower: that AI agents can notice something odd in raw data and chase it, as a human expert would.[1][2]

How the agents searched the DNA
Anthropic says its involvement was limited to the opening prompt and the later lab work. Roughly 950 Claude agents spent 21 hours and 210 million tokens combing the data. They gathered more than 200,000 reverse transcriptases, enzymes that copy RNA into DNA, picked out 3,500 new candidate systems and wrote reports on the 20 most promising.[1]
The preprint, by six Anthropic researchers, says the agents ran as Claude Code instances and surveyed reverse transcriptase regions across 1.9 billion protein clusters from metagenomic databases. Anthropic says this kind of analysis can take an expert scientist weeks to months.[2][1]

What the ART system is, and what is still unknown
The family, named array-associated reverse transcriptases or ART, is found mainly in bacteriophages, the viruses that infect bacteria. It has three parts: the enzyme, a partner gene beside it and a long array of evenly spaced repeats of about 200 DNA letters each. The enzyme itself had been noted before. Anthropic says Claude appears to be the first to notice the repeats and the partner gene.[1][2]
Early experiments show the array is read out as a set of distinct short RNAs during infection by a Staphylococcus phage. That hints the enzyme may be steered by many different RNA guides, much as CRISPR is. Anthropic stresses that the system's function is still being worked out, and that similar systems are programmable tools that cut, copy or paste DNA.[2][1]
Human scientists still run the lab
Anthropic says its Bay Area lab looks like any molecular biology lab, works only at the lowest biosafety levels, BSL-1 and BSL-2, and handles no pathogens that infect humans. All bench work is done by human scientists, and candidates are tested only after passing the team's own review.[1]
Feng Zhang, a CRISPR pioneer at MIT and the Broad Institute, called the finding genuinely intriguing after reviewing the preprint and said it merits further investigation. The tools used are Claude Science and Claude Code, which Anthropic notes are available to any scientist. Six days earlier it opened a verification programme giving vetted life science professionals more permissive biology safeguards.[1][3]
What it could mean for Malaysian genomics
Malaysia already produces the kind of data this approach targets. The Malaysia Genome and Vaccine Institute, run by the National Institutes of Biotechnology Malaysia since January 2012, studies tropical bioresources through genome sequencing, comparative genomics and structural biology, with its own high-throughput sequencing and bioinformatics platforms.[4]
The MyGenom project plans to sequence 2,400 genomes from healthy Malaysians in its first phase and 10,000 in the second, under a flagship programme of the National Biotechnology Policy 2.0. In Sarawak, the Biodiversity Council agreed in October 2025 to link its biodiversity databases with AI. None of these bodies is connected to Anthropic's work.[5][6]
Why Malaysia should care
Malaysian institutes already sequence tropical bioresources and Malaysian genomes. Anthropic's result suggests AI agents could help sift such data, but no Malaysian group is involved and the finding is unconfirmed.
University researchers
Agents can shortlist leads in public data.[1]
Practical move: Pilot one on an existing sequence dataset.
Biotech firms
Faster early screening, same lab costs.[1][2]
Practical move: Budget for wet-lab checks of AI leads.
Research funders
Compute and data now shape discovery.[2]
Practical move: Fund data curation alongside compute.
What Malaysians can do now
- Research groups can try an AI agent on a dataset they already know well and check whether it finds what their own curators found before trusting it on new data.
- Treat any AI-found candidate as a hypothesis until bench experiments confirm it, and keep a record of the prompts and data behind each lead.
- Watch for peer review of the ART preprint before citing its claims as settled science.
What we still do not know
What we still do not know
- What the ART system actually does inside a cell.
- Whether peer reviewers will accept the preprint's findings.
- How often agent-generated leads fail in the lab.
Sources
- 1.Claude discovers a novel enzyme system Anthropic, 23 September 2026
- 2.Autonomous AI agents discover reverse transcriptases with tandem repeat arrays Anthropic, 23 September 2026
- 3.Introducing the Life Sciences Verification Program Anthropic, 17 September 2026
- 4.MGVI – National Institutes of Biotechnology Malaysia National Institutes of Biotechnology Malaysia
- 5.MyGenom | Malaysia Genome Project | A study to establish a Malaysia reference genomic dataset MyGenom, 2 September 2024
- 6.Memorandum of Understanding between Sarawak Biodiversity Council and Artificial Intelligence Centre (SAIC) Sdn Bhd Sarawak Biodiversity Centre, 22 October 2025
- 7.File:APHIS-Moore Air Base in Texas-Plant Protection and Quarantine Science and Technology Insect Management and Molecular Diagnostics Laboratory (20230621-CDP-APHIS-0556).jpg Wikimedia Commons, 21 June 2023
- 8.File:Enterobacteria phage T2 transmission electron micrograph.jpg Wikimedia Commons, 14 January 2022
- 9.File:Bangunan Makmal Fakulti Sains dan Teknologi UKM.jpg Wikimedia Commons, 9 December 2023


