{"slug": "claude-finds-hidden-dna-system-with-crispr-like-features-that-could-cut-copy-and", "title": "Claude Finds Hidden DNA System With CRISPR-Like Features That Could Cut, Copy and Paste Genetic Material", "summary": "Anthropic announced on 23 September that its Claude AI agents screened more than 200,000 reverse transcriptases and identified a previously uncharacterised enzyme system in bacteriophage DNA, named array-associated reverse transcriptases (ART), which has a layout resembling CRISPR arrays. Roughly 950 Claude agents spent about 21 hours and 210 million tokens analysing the data, narrowing approximately 3,500 candidate systems to 20 compelling candidates before Anthropic's scientists investigated the candidate in the laboratory. MIT professor and CRISPR genome-editing pioneer Feng Zhang, who reviewed the preprint, called the RNA-repeat arrays associated with reverse transcriptases 'genuinely intriguing' and said they merit further investigation, though Anthropic cautioned that ART's function and any practical medical use remain unproven.", "body_md": "# Claude Finds Hidden DNA System With CRISPR-Like Features That Could Cut, Copy and Paste Genetic Material\n\n## AI agents screened 200,000+ reverse transcriptases, uncovering array-associated enzymes with unknown applications\n\nAnthropic's Claude has discovered a previously uncharacterised enzyme system hidden in bacteriophage DNA, after AI agents searched more than 200,000 reverse transcriptases and spotted an unusual genetic pattern resembling CRISPR.\n\nThe system, named array-associated reverse transcriptases (ART), could potentially represent a new programmable biological mechanism. But scientists do not yet know exactly what it does, whether it can edit DNA or whether it will have any practical medical use.\n\n## Claude Spotted What Humans Had Missed\n\nAnthropic announced the discovery on 23 September as the first result from its new molecular biology research group and laboratory.\n\nResearchers initially gave Claude broad instructions to search a massive database of DNA sequences for interesting examples of reverse transcriptases, enzymes that copy RNA into DNA.\n\nRoughly 950 Claude agents then spent around 21 hours analysing the data, using 210 million tokens. They gathered more than 200,000 reverse transcriptases, identified approximately 3,500 candidate systems, and eventually narrowed them to 20 particularly compelling candidates.\n\nWhile examining one unusual reverse transcriptase family, Claude inspected the surrounding raw DNA and detected a repeating pattern beside the gene.\n\nThe AI counted the repeats, examined their spacing, compared the arrangement with known systems and searched scientific literature for previous descriptions before flagging it for human review.\n\nAnthropic's scientists subsequently investigated the candidate in their laboratory.\n\n## Why Scientists Are Comparing It to CRISPR\n\nART consists of three main components: a reverse transcriptase, a neighbouring partner gene and a long array of evenly spaced DNA repeats.\n\nThat layout resembles CRISPR arrays, which helped turn a naturally occurring bacterial defence system into one of biotechnology's most important gene-editing tools.\n\nAnthropic said the characteristics found in ART have previously appeared together in only a handful of known biological systems, all of which are programmable and perform operations such as cutting, copying or pasting DNA.\n\nThat does not mean ART itself has been shown to perform those functions.\n\nInitial experiments found that its DNA array is expressed as distinct short RNAs, which Anthropic said suggests something analogous to other programmable systems could be occurring. Further experiments are underway to establish ART's primary function.\n\n## Could It Become Another Gene-Editing Tool?\n\nAnthropic is cautious about the discovery's significance.\n\nThe underlying reverse transcriptase had previously been identified in a jumbo phage. What Claude appears to have discovered was the broader biological system surrounding it, including the non-coding repeat array and an additional protein with an unknown function.\n\nFeng Zhang, an MIT professor and CRISPR genome-editing pioneer who reviewed the preprint, called the RNA-repeat arrays associated with reverse transcriptases 'genuinely intriguing' and said they merit further investigation.\n\nThe research remains early-stage, and Anthropic has released its findings as a preprint while work continues.\n\nFor scientists, the immediate significance may extend beyond whether ART eventually becomes a useful genetic tool. Claude's ability to sift through enormous genomic datasets and identify an overlooked biological pattern suggests AI could increasingly help researchers find promising discoveries hidden in data that would take humans weeks or months to analyse.\n\nThe discovery could also provide an important test of how effectively AI can contribute to future experimental research alongside human scientists.\n\n© Copyright IBTimes 2026. All rights reserved.", "url": "https://wpnews.pro/news/claude-finds-hidden-dna-system-with-crispr-like-features-that-could-cut-copy-and", "canonical_source": "https://www.ibtimes.co.uk/anthropic-claude-discovers-crispr-like-enzyme-system-1821665", "published_at": "2026-09-24 08:47:22+00:00", "updated_at": "2026-09-24 09:32:32.079821+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-research", "generative-ai"], "entities": ["Anthropic", "Claude", "Feng Zhang", "MIT", "array-associated reverse transcriptases (ART)", "CRISPR", "IBTimes"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/claude-finds-hidden-dna-system-with-crispr-like-features-that-could-cut-copy-and", "markdown": "https://wpnews.pro/news/claude-finds-hidden-dna-system-with-crispr-like-features-that-could-cut-copy-and.md", "text": "https://wpnews.pro/news/claude-finds-hidden-dna-system-with-crispr-like-features-that-could-cut-copy-and.txt", "jsonld": "https://wpnews.pro/news/claude-finds-hidden-dna-system-with-crispr-like-features-that-could-cut-copy-and.jsonld"}}