{"slug": "anthropic-s-claude-found-a-new-enzyme-system-that-looks-like-crispr", "title": "Anthropic's Claude Found a New Enzyme System That Looks Like CRISPR", "summary": "Anthropic reported on September 23, 2026 that its Claude model ran roughly 950 agents for 21 hours and about 210 million tokens over public DNA databases, flagging a previously unknown bacteriophage enzyme system the company calls ART (array-associated reverse transcriptases) that structurally resembles a CRISPR array. Anthropic's Bay Area lab, which operates only at biosafety levels 1 and 2, ran the wet-lab experiments showing the repeat array produces short RNAs, while MIT and Broad Institute professor Feng Zhang, who reviewed the preprint, called the RNA-repeat arrays \"genuinely intriguing\" and said they merit further investigation. Anthropic CEO Dario Amodei noted Stanford researchers previously described a system in some ways similar, tempering the claim of a wholly unprecedented discovery, and the company says it does not yet know what ART does.", "body_md": "*Anthropic says its Claude model spent 21 hours combing viral DNA and flagged an unknown enzyme system that structurally echoes CRISPR. Feng Zhang, the scientist who helped turn CRISPR into a gene-editing tool, calls the find \"genuinely intriguing.\"*\n\nClaude was sorting through raw bacteriophage DNA. Then it stopped. What it typed next reads more like a lab notebook than a chatbot log. \"I can see by eye a tandem repeat array ... that's a CRISPR-like ... repeat array?!\" the model wrote, according to Anthropic's own account of the moment, posted in a blog and an accompanying preprint on September 23, 2026. That single flagged sequence is now the centerpiece of what Anthropic is calling one of the first real scientific discoveries credited to an AI model working largely on its own.\n\nThe system doesn't have a catchy name yet. Anthropic is calling it ART, short for array-associated reverse transcriptases. It shows up in bacteriophages, the viruses that infect bacteria, and it has three parts: a reverse transcriptase enzyme, which copies RNA back into DNA, a neighboring partner gene, and a long run of evenly spaced DNA repeats arranged in a pattern that looks a lot like a CRISPR array. Anthropic says early lab experiments already show that repeat array producing short RNAs, which is exactly what you'd expect if the whole system is built to guide some kind of DNA-cutting or DNA-editing machinery. Nobody at Anthropic actually knows what ART does yet. That's the honest, slightly awkward part of this story that tends to get flattened in the headlines.\n\nThis wasn't one clean prompt and one clean answer. Anthropic's life sciences group, which it only stood up in spring 2026, ran roughly 950 Claude agents over 21 hours, burning through about 210 million tokens combing public DNA databases. The agents pulled together more than 200,000 known and candidate reverse transcriptases, narrowed that pile to 3,500 unusual ones, and then narrowed again to 20 genuinely compelling candidates worth a closer look. Human scientists in Anthropic's Bay Area lab, which operates only at biosafety levels 1 and 2 and handles no pathogens that infect people, ran the actual wet-lab experiments. Claude did the search. People did the science. Anthropic is careful to draw that line right now, and it should be.\n\nFeng Zhang reviewed the preprint before it went public. He's not a bystander here: he's the MIT and Broad Institute professor whose lab turned CRISPR from a bacterial curiosity into the gene-editing tool that's since won a Nobel Prize and spawned an entire industry. His verdict was measured, not breathless. He called the RNA-repeat arrays Claude flagged \"genuinely intriguing\" and said they merit further investigation. That's a scientist's way of saying: interesting lead, not yet a result.\n\n[Anthropic Says It Blocked Scientists Using Claude for Bioweapons Research](https://startupfortune.com/anthropic-says-it-blocked-scientists-using-claude-for-bioweapons-research/)\n\nAnthropic disclosed five cases where researchers tried to use Claude for research that could aid biological weapons development, including a May 2026 attempt tied to a military institute studying the chikungunya virus. The same September 2026 threat report details a Russian state-linked hacking operation against Ukraine and drone-swarm software... - [how Claude was used for bioweapons research](https://startupfortune.com/anthropic-says-it-blocked-scientists-using-claude-for-bioweapons-research/) - [Anthropic blocks scientists from gain of function research](https://startupfortune.com/anthropic-says-it-blocked-scientists-using-claude-for-bioweapons-research/)\n\nDario Amodei has his own caveat, and it's worth taking seriously. Anthropic's CEO has acknowledged that Stanford researchers previously described a system in some ways similar to what Claude found, which tempers the claim that this is a wholly unprecedented discovery. He's also said the quiet part out loud: \"Eventually it may even be possible for Claude itself to safely perform the experiments ... but we aren't doing that today.\" Read that sentence twice. It's a roadmap, not just a disclaimer.\n\nHere's the thing that makes this bigger than one lab result. If a frontier AI lab can point its own model at public genomic data and surface a patentable-looking biological mechanism, that's a new kind of competitive asset sitting next to chips and model weights. It also raises the exact question biosecurity researchers have been asking for two years: what happens when the same search process that finds a useful gene-editing system can just as easily be pointed at something dangerous. Anthropic's answer, for now, is to keep humans running every experiment and keep the lab at the lowest biosafety tiers. That's a reasonable line to hold. Whether every lab racing to copy this approach holds it too is a different question entirely.\n\nThe market noticed fast, and not kindly toward the incumbents. CRISPR Therapeutics shares fell nearly 6% on September 23, slipping to around $56.39 in afternoon trading, according to GuruFocus. Other gene-editing stocks, including Intellia, Beam and Editas, slipped the same day. That's a jumpy reaction to a preprint that hasn't been peer reviewed, describing a system nobody yet understands the function of. But jumpy is what markets do when the word CRISPR and the word AI show up in the same sentence.\n\nThe preprint is public. The function of ART is not yet known. Anthropic says more experiments are already underway.\n\n**Also read:** [Huawei and Cambricon Are Charging More for AI Chips as Memory Runs Short](https://startupfortune.com/huawei-and-cambricon-are-charging-more-for-ai-chips-as-memory-runs-short/) • [Tom Siegel, Who Built Google's Safety Team, Warns AI Is Failing Kids](https://startupfortune.com/tom-siegel-who-built-googles-safety-team-warns-ai-is-failing-kids/) • [Nvidia's Revenue Forecast Assumes China Buys Zero AI Chips This Quarter](https://startupfortune.com/nvidias-revenue-forecast-assumes-china-buys-zero-ai-chips-this-quarter/)\n\n*This article is posted in [AI News](https://startupfortune.com/category/ai/), check it out for more related stories.*\n\n[This Week in AI: Robots Take a Punch, Assistants Get Long-Term Memory, and Models Keep Getting Cheaper](https://startupfortune.com/this-week-in-ai-robots-take-a-punch-assistants-get-long-term-memory/)\n\nA dense week across robotics, model releases and AI research: humanoid robots duke it out, ChatGPT gets long-term memory, and frontier-adjacent AI keeps getting cheaper and faster. - [how to build robots that handle physical impacts](https://startupfortune.com/this-week-in-ai-robots-take-a-punch-assistants-get-long-term-memory/) - [AI models getting cheaper and more accessible features](https://startupfortune.com/this-week-in-ai-robots-take-a-punch-assistants-get-long-term-memory/)\n\n## Join the discussion\n\n[Open in the community →](https://startupfortune.com/community/)\n\nAlmost there. Sign in and your reply posts straight away.", "url": "https://wpnews.pro/news/anthropic-s-claude-found-a-new-enzyme-system-that-looks-like-crispr", "canonical_source": "https://startupfortune.com/anthropics-claude-found-a-new-enzyme-system-that-looks-like-crispr/", "published_at": "2026-09-24 00:11:59+00:00", "updated_at": "2026-09-24 00:29:02.134726+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-research", "large-language-models", "ai-agents", "ai-safety"], "entities": ["Anthropic", "Claude", "Feng Zhang", "MIT", "Broad Institute", "Dario Amodei", "ART", "Stanford"], "alternates": {"html": "https://wpnews.pro/news/anthropic-s-claude-found-a-new-enzyme-system-that-looks-like-crispr", "markdown": "https://wpnews.pro/news/anthropic-s-claude-found-a-new-enzyme-system-that-looks-like-crispr.md", "text": "https://wpnews.pro/news/anthropic-s-claude-found-a-new-enzyme-system-that-looks-like-crispr.txt", "jsonld": "https://wpnews.pro/news/anthropic-s-claude-found-a-new-enzyme-system-that-looks-like-crispr.jsonld"}}