The news
Anthropic said on September 23 that its new life sciences lab used Claude to identify a previously uncharacterized enzyme system in viruses that infect bacteria. The Claude enzyme discovery has features Anthropic says are reminiscent of CRISPR, the gene-editing technology.
According to Anthropic, scientists gave Claude an initial prompt to search a large DNA sequence database for new reverse transcriptases, enzymes that copy RNA into DNA. Roughly 950 Claude agents worked for 21 hours and used 210 million tokens, gathering more than 200,000 reverse transcriptases, identifying 3,500 new candidate systems and narrowing them to 20 for detailed reports. Anthropic said this kind of analysis can take an expert weeks to months.
One agent noticed a repeating DNA pattern next to an unusual reverse transcriptase gene. The resulting system, which Anthropic calls ART, consists of the enzyme, a partner gene and a long array of evenly spaced DNA repeats resembling a CRISPR array. Anthropic said the underlying enzyme had been identified in earlier studies, but that Claude appears to be the first to notice the system's defining features. Early experiments show the array is expressed as distinct short RNAs; what the system does is still unknown, and a pre-print has been released.
The lab work was done by people. Anthropic said its Bay Area lab handles only lower biosafety levels, BSL-1 and BSL-2, does not work with pathogens that infect humans, and that all experiments were performed by human scientists. The team uses Claude Science and Claude Code, which Anthropic noted are available to any scientist. CRISPR pioneer Feng Zhang of MIT and the Broad Institute reviewed the pre-print and praised it as a sign that AI agents can contribute to biology research, Anthropic said. The group sits in Anthropic's broader life sciences organization, which the company said also includes teams working on drug discovery and on training Claude in biology and chemistry.
TechCrunch reported that CEO Dario Amodei acknowledged the work built on others' research and that a Stanford team had previously found a system that is in some ways similar. Amodei wrote that the discovery was made mostly, though not entirely, by Claude, TechCrunch said.
The numbers
- Claude agents deployed
- About 950
- Search time
- 21 hours
- Tokens used
- 210 million
- Reverse transcriptases gathered
- 200,000+
- Candidates narrowed
- 3,500 to 20
Why CEOs should care
For pharma and biotech R&D leaders, the notable number is not the discovery but the throughput: a day of parallel agent work to survey a protein family that Anthropic says would take a specialist weeks or months. Heads of discovery should run a controlled pilot on a well-understood problem, compare agent output with past human work, and measure how many agent-generated hypotheses survive expert review and lab testing. Ask any vendor whether agent runs can be reproduced and whether the evidence behind each candidate is logged for scientists to review.
CFOs should budget for the full pipeline, not just compute. Anthropic said its scientists' involvement was limited to the initial prompt and the lab work, with agents narrowing 3,500 candidates to 20 detailed reports, but finding out what any candidate actually does still depends on human-run wet-lab experiments. The cost savings come from narrowing what humans test, so the business case rests on hit rates and the capacity of lab teams downstream.
Chief risk officers and boards should note the safeguards Anthropic describes: low-biosafety work only, human-run experiments and a separate verification program that gives life-science professionals access to models with safeguards adjusted for biology work. Any company adopting AI in biology should document similar controls, including who can run which agents on which data, and how outputs involving hazardous biology are screened.
The bigger picture
AI developers are moving from supplying tools to scientists toward doing science themselves. Anthropic has joined a field that TechCrunch noted already includes AI-assisted enzyme design and protein-structure tools, and the company says it wants outside scientists to propose research questions. If agent-led hypothesis generation holds up under peer review, the bottleneck in early discovery shifts from finding candidates to testing them, which favors organizations with strong lab operations and good data.
What’s next
Watch for peer review of the pre-print, follow-up experiments on what the ART system actually does, and any partnerships Anthropic strikes with drug developers or academic labs. According to TechCrunch, Amodei has said Claude might eventually run experiments itself by controlling lab equipment, with appropriate safeguards, but that Anthropic is not doing that now.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made with our newsroom’s technology tools, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is not published automatically; it is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
Companies in this story
