single research brief<\/a>. Find new reverse transcriptase systems in a database of 1.9 billion protein clusters. Reverse transcriptases are enzymes that copy RNA into DNA.<\/p>\n\n\n\nAgents running Claude Mythos 5 did the rest. One agent planned and ran each task. A second agent reviewed the work. Agents opened new tasks as they went.<\/p>\n\n\n\n
The scale was large. The campaign ran for about 21 hours without human input. It used roughly 950 agent sessions and more than 210 million tokens. The agents gathered over 200,000 reverse transcriptases. They flagged 3,500 candidate systems. They then narrowed the list to the 20 most compelling and wrote reports for human review.<\/p>\n\n\n\n
Anthropic notes that this kind of analysis can take an expert scientist weeks or months.<\/p>\n\n\n\n
The Moment of Recognition<\/h4>\n\n\n\n
The heart of the Claude enzyme discovery came from a side path. One agent was reading raw DNA next to an unusual enzyme. It wrote that it could “see by eye” a tandem repeat array. It then behaved like a human scientist. It counted the repeats. It measured their spacing. It compared the pattern with known systems. It searched the scientific literature. Only then did it file a report.<\/p>\n\n\n\n
The enzyme itself was not new. An earlier study had found it in a jumbo phage. That first genome report described the enzyme but not the repeats or the partner gene. Claude appears to be the first to notice the full system around it.<\/p>\n\n\n\n
What the System Looks Like<\/h4>\n\n\n\n
The system at the center of the Claude enzyme discovery is called array-associated reverse transcriptase, or ART. It has three parts. There is a reverse transcriptase. There is a partner protein with no known function. And there is a long array of evenly spaced DNA repeats, laid out much like a CRISPR array.<\/p>\n\n\n\n
Two details stand out. None of the systems carries the cas genes that CRISPR systems keep nearby. And ART turns up mainly in phages, while CRISPR sits mostly in bacteria.<\/p>\n\n\n\n
Early lab work adds a clue. The repeat array is read out as several distinct short RNAs. In published data from a Staphylococcus phage, these RNAs made up as much as 8 percent of the phage’s RNA 15 minutes after infection. This matters because CRISPR arrays also produce short RNAs. Those RNAs make CRISPR programmable.<\/p>\n\n\n\n
What Has Not Been Shown<\/h4>\n\n\n\n
The Claude enzyme discovery is not yet a working tool. Anthropic states the caveats plainly. The preprint says the team has not yet shown that the enzyme is active. It has not shown that the enzyme works on these RNAs. The function of the system is unknown.<\/p>\n\n\n\n
Anthropic also says only a handful of known systems share this set of features. All of them can cut, copy, or paste DNA. That is a reason for interest. It is not evidence that ART can do the same.<\/p>\n\n\n\n
What Experts Are Saying<\/h4>\n\n\n\n
Reactions to the Claude enzyme discovery were mixed but mostly measured. Feng Zhang, a CRISPR pioneer at MIT and the Broad Institute, reviewed the preprint. He called the RNA-repeat arrays “genuinely intriguing” and said they merit further investigation. That is real praise, but it is careful praise.<\/p>\n\n\n\n
Outside voices were more cautious. One expert, Blake, told Al Jazeera that many people had leaped to conclude Anthropic found the next Nobel-winning gene editor. He noted that CRISPR in nature is a bacterial immune system. That is very different from CRISPR the technology. He also said countless CRISPR-like sequences remain uncatalogued because millions of bacterial species are still unstudied.<\/p>\n\n\n\n
Both views can be true. The system is worth studying. It is not yet a tool.<\/p>\n\n\n\n
What Needs to Happen Next<\/h4>\n\n\n\n
Turning the Claude enzyme discovery into established science takes several steps.<\/p>\n\n\n\n
First, researchers must test whether the enzyme is active. This means purifying it and running biochemical assays.<\/p>\n\n\n\n
Second, they must learn what the short RNAs do. They must find out whether the partner protein is required.<\/p>\n\n\n\n
Third, they must identify what the system targets. That could be phage RNA, bacterial DNA, or something else.<\/p>\n\n\n\n
Fourth, independent labs must reproduce the results. Until then, this remains one group’s preprint.<\/p>\n\n\n\n
Only after these steps can anyone ask whether the system could be engineered for editing.<\/p>\n\n\n\n
Potential Uses, Kept in Proportion<\/h4>\n\n\n\n
If ART proves to be programmable, the Claude enzyme discovery could add a new tool to the gene editing toolkit. Different systems often have different sizes and strengths. That can suit different medical needs. Anthropic itself says it will take much more work to learn what this system does and whether it can be put to similar use.<\/p>\n\n\n\n
There is a second, broader use. The method behind the Claude enzyme discovery matters as much as the molecule. Agents mined a huge DNA dataset and surfaced a pattern that human researchers had missed. That approach could be applied to many other enzyme families. It also fits a wider trend covered on this blog. AI systems are moving from summarizing science to contributing to it.<\/p>\n\n\n\n
The Human Role Remains Central<\/h4>\n\n\n\n
The Claude enzyme discovery also shows where humans stayed involved. People wrote the research brief. People reviewed the final reports. People ran the wet lab experiments that showed the repeat array produces distinct RNAs. The agents found a lead. They did not prove anything about biology.<\/p>\n\n\n\n
This division of labor is likely to define AI-assisted discovery for now. Agents can search at a scale no team can match. Scientists must still confirm what the search finds.<\/p>\n\n\n\n
Conclusion<\/h4>\n\n\n\n
The Claude enzyme discovery is a genuine result, but a narrow one. An AI agent<\/a> spotted a CRISPR-like repeat array beside a known enzyme and flagged a real, previously unnoticed system. Early experiments show the array is expressed as distinct short RNAs. What ART does, and whether it can be used, remains unknown.<\/p>\n\n\n\nThe right reading sits between hype and dismissal. This is not the next CRISPR. It is a credible lead, found at machine speed, that now needs slow and careful human science. If ART turns out to be programmable, the Claude enzyme discovery will be remembered as an early example of AI-assisted biology paying off. If it does not, the method will still have shown its value.<\/p>\n","protected":false},"excerpt":{"rendered":"
A Discovery Worth Understanding Carefully Last week, Anthropic announced that Claude had found a previously unknown enzyme system in the DNA of bacteriophages, the viruses that infect bacteria. The Claude enzyme discovery drew fast headlines. Many of them ran ahead of the facts. Some called it the next CRISPR. Anthropic itself was far more careful. This article explains the Claude enzyme discovery in plain steps. It covers how the work was done, what was found, and what still needs to be proven.<\/p>\n","protected":false},"author":1,"featured_media":1401,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3],"tags":[],"class_list":["post-1400","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-engineering"],"_links":{"self":[{"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/posts\/1400","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/comments?post=1400"}],"version-history":[{"count":1,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/posts\/1400\/revisions"}],"predecessor-version":[{"id":1402,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/posts\/1400\/revisions\/1402"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/media\/1401"}],"wp:attachment":[{"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/media?parent=1400"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/categories?post=1400"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/learnerbox.net\/blog\/wp-json\/wp\/v2\/tags?post=1400"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}