Anthropic's 950 AI Agents Just Found a Hidden Biological System
Hundreds of AI agents did months of research in days — and found something new.
Anthropic, the company behind the Claude chatbot, pointed roughly 950 copies of its AI at a massive biology question: sorting through more than 200,000 reverse transcriptases, a type of enzyme. One agent noticed something odd — a long, evenly spaced repeating pattern of DNA sitting next to an enzyme gene. The AI checked the spacing, compared it to known biological systems, searched scientific literature for prior reports, and flagged it. Human researchers then took that lead into the lab, confirmed the repeats are copied into short RNA molecules, and named the new system array-associated reverse transcriptases, or ART. It shows up mainly in bacteriophages, viruses that infect bacteria.
The ART find is drawing attention because CRISPR, the tool that made gene editing possible, also relies on arrays of repeated DNA sequences. That resemblance is intriguing, not proof. Similar structure does not mean similar function, and researchers still have no idea what ART does or whether it has any practical use. Feng Zhang, a pioneer of CRISPR at MIT, called the link between RNA-repeat arrays and reverse transcriptases worth investigating further. The findings also haven't been reviewed by other scientists yet.
The bigger story is how the discovery happened. Rather than asking the AI to summarize existing papers, Anthropic gave it an open-ended research goal. Hundreds of agents searched databases, investigated enzyme families, eliminated weak candidates, and wrote up reports on the most promising leads. Humans were still the ones deciding what mattered and running the experiments. Think of it as a tireless research assistant that reads millions of pages overnight and hands you a shortlist — you still do the thinking and the testing.
That matters because early-stage research is expensive and slow. Every candidate a lab chases costs money, equipment, and skilled people. If AI can shrink a list of 200,000 possibilities to a handful worth testing, companies and universities save time and cash before experiments begin. This isn't AI curing diseases on its own. It's AI making the search phase faster, which could mean new medicines and lab tools arrive sooner — and cheaper — than they otherwise would.
- Anthropic used about 950 Claude AI agents to scan over 200,000 enzymes and flag one unusual pattern humans had never described.
- The new system, called ART, has repeating DNA similar to CRISPR — but scientists don't yet know what it does or if it's useful.
- The practical win is speed and savings: AI narrows thousands of candidates so costly lab experiments focus only on the best leads.
Why It Matters
AI can now do the tedious searching part of science, so new medicines and tools may arrive faster and cheaper.