Anthropic's first wet-lab result: Claude autonomously finds a CRISPR-like enzyme system it calls ART

Start with a piece of background the company itself wrote, because it shapes how the finding should be read. Anthropic says that in the spring of 2026 it formed a research group to see whether general AI models can systematize and accelerate such discoveries, which required building its own lab and having a single team work on everything from training Claude in biology to running experiments in the lab. In other words, this is not a model offering an idea in a chat window, but a full chain from data search to wet-lab validation, built specifically for this purpose.

The second thread is the boundary of the discovery, the part most easily overstated and therefore most important to pin down. Per the official post, the system is based on a reverse transcriptase (RT), an enzyme that copies RNA into DNA, and that underlying RT "had been identified in previous studies." Claude's contribution is that it appears to be the first to notice the system's defining features: an associated array of non-coding DNA sequences and an additional accessory protein of unknown function. The accurate framing is that Claude noticed a combination of features others had missed and completed the picture, not that it invented a new enzyme from scratch. The official phrasing is that "Claude appears to be the first to notice."

The third thread is how it was executed, because it offers a workflow that can actually be discussed. Per the official post, the human contribution was a single prompt: search a massive database of DNA sequences for interesting new examples of RTs. Anthropic states plainly that its involvement was limited to the initial prompt and the lab work, while Claude agents combed through the database, investigated distinct RT families, and used their own judgment to identify interesting candidates. The numbers: agents gathered over 200,000 RTs, picked out 3,500 new candidate systems, and narrowed those to the 20 most compelling candidates; after roughly 950 agents spent 21 hours and 210 million tokens searching, one of them spotted something remarkable, a repeating pattern of DNA sequences occurring next to the gene for an odd-looking RT. After further analysis and lab testing, they recognized this pattern marked a previously uncharacterized enzyme system found in bacteriophages, the viruses that infect bacteria.

The fourth thread is external evaluation, and one name here matters. Per the official post, after reviewing the pre-print, Feng Zhang, one of the pioneers of CRISPR genome editing and a professor at MIT and the Broad Institute, commented: "This is an exciting example of how AI agents can contribute to biological discovery. The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigation. I hope this work encourages more scientists to explore how AI can support their research." Having a central figure in CRISPR comment on a CRISPR-like finding suggests Anthropic wants the result received by the molecular biology community, not just read as an AI demo.

The fifth thread is the lab itself, because that is where the official post narrows the most alarming possible reading of "AI doing science." Per the official post, the lab is located in the Bay Area and looks like a typical molecular biology lab; the research involves only the lower levels of the biosafety risk level, BSL-1 and BSL-2, and does not handle pathogens that can infect humans. It states plainly: "All of the lab work is performed by human scientists." The post notes they have experimented with using AI to accelerate lab work, in initiatives like the Model Hardware Standard, but found that approach less conducive to the ad hoc workflows involved in molecular biology research. Together, these points mean the AI handles pattern-finding and hypothesis generation at scale, while the hands at the bench remain human.

The sixth thread is third-party reading of the news, which should be kept separate from the official account. Per TechCrunch on September 23, Anthropic says it found a previously unknown enzyme system hidden in the DNA of bacteriophages, behaving similarly to CRISPR in that it can "perform operations like cutting, copying, and pasting DNA"; the report also notes it will be up to the broader research community to validate how big, or new, this discovery actually is. TechCrunch adds that Anthropic CEO Dario Amodei acknowledges the discovery was based on the previous work of others, and mentions that a team from Stanford previously discovered a related system. Per The Verge on September 24, this is the first result from Anthropic's newly launched wet lab and an early test of Claude's usefulness for science as the company prepares to go public. Those two items are media judgments; the official post contains no statement about going public.