Anthropic's AI Found a DNA-Editing Tool in Just 21 Hours

Anthropic, the company behind the Claude chatbot, quietly opened a real biology lab in the Bay Area this spring. Instead of just testing Claude on text, the company pointed it at genetics data. Working with roughly 950 "agents" — think of them as hundreds of AI assistants running at once — Claude scanned a mountain of genetic material and flagged something nobody had logged before: a hidden enzyme system inside viruses that infect bacteria. The whole search took 21 hours. A human scientist might spend years on the same hunt.

According to Anthropic, the system behaves like CRISPR, the gene-editing tool that won a Nobel Prize and now powers treatments for sickle cell disease. Like CRISPR, it appears able to cut, copy and paste DNA — the basic editing moves behind modern genetic medicine. New tools like this could speed up drug development and, eventually, help treat diseases that are currently incurable. Anthropic's CEO, Dario Amodei, has said he believes AI could "cure most diseases in 5-10 years."

The catch is that this isn't purely an AI win. Stanford researchers previously found a similar system, and Anthropic admits Claude built on other people's work. Outside scientists still need to verify how new or useful the discovery really is. Also, humans — not Claude — performed every physical experiment. The lab handles only low-risk materials and no germs that infect people.

There's a bigger tension too. The same AI leaders, including Amodei, have publicly warned that AI is powerful enough to be dangerous, and he has named bioweapons as a top fear. Yet his company is racing ahead in biology, arguing the benefits outweigh the risks. Amodei says a future lab might let Claude control equipment on its own, but "we aren't doing that today." AI in biology is already mainstream: Google's AlphaFold arrived back in 2020.