Anthropic announced that its AI model Claude has autonomously discovered a new enzyme system in bacteriophage DNA that exhibits structural features reminiscent of CRISPR. The system, named ART (array-associated reverse transcriptases), was flagged after Claude spent 21 hours combing through a large database of DNA sequences. If validated, this could become the first genuine scientific discovery credited to an AI operating largely on its own.

What Happened

Anthropic CEO Dario Amodei revealed on September 24 that Claude—prompted by researchers at the company’s biology lab in San Francisco—identified a tandem repeat array alongside a reverse transcriptase that “shows a pattern reminiscent of CRISPR.” The system was found in bacteriophage DNA, the viruses that infect bacteria. “We don’t yet understand what this system does, but only a handful of known systems share its features, and all of them are able to cut, copy, and paste DNA,” Anthropic stated in a blog post.

The discovery is significant because CRISPR technology has revolutionized gene editing, leading to treatments for sickle cell disease and cancer. CRISPR pioneer Feng Zhang called the find “genuinely intriguing” but cautioned that the system’s function remains unproven. Shares of CRISPR Therapeutics dropped more than 4% on the news, reflecting both excitement and uncertainty. Anthropic published a preprint alongside its announcement, detailing the AI’s raw output, which read like a lab notebook: “I can see by eye a tandem repeat array … that’s a CRISPR-like … repeat array?!”

Read the full announcement →

My Take

This is a landmark moment, but not for the reason most headlines suggest. Yes, a potential new gene-editing tool is huge. What’s bigger is that Claude likely brute‑forced the discovery using roughly 950 AI agents to sift through viral DNA—something a human PhD student might have taken years to do. That signals an exponential shift in biology research: AI isn’t just helping write code or chat; it’s now generating testable hypotheses that could lead to new therapeutics.

For developers and data scientists, this raises the bar for what “AI‑assisted discovery” means. Anthropic’s model didn’t just predict a structure—it recognized a pattern that humans overlooked, asked the right questions, and produced a coherent report. Expect more labs to adopt similar AI‑driven approaches, especially in drug target identification and enzyme engineering. The real value won’t be in one discovery but in the pipeline it enables.

What to Watch

  • Validation experiments: If ART works as a programmable DNA‑editing system, expect a flurry of biotech startups and patent filings within months.
  • Regulatory reactions: CRISPR is already controversial; a new, AI‑discovered tool will likely trigger fresh debates on oversight and ethical use.
  • Competitor moves: DeepMind’s AlphaFold and Meta’s ESM‑2 already excel at protein structure prediction. This discovery proves frontier models like Claude can now discover biological systems from scratch, intensifying the race for scientific AI.