Anthropic Says It Discovered a Crispr-Like System. Now What?

WIRED reports Anthropic’s September 23 claim that Claude identified an enzyme system “reminiscent of Crispr.” About 950 agents searched genomic databases for novel reverse transcriptases, narrowing to an “unusual” family with repeat DNA arrays Anthropic calls ART (array-associated reverse transcriptases) in jumbo phages. One physical experiment appears in a non-peer-reviewed technical report; the company says it does not yet understand what the system does.

Experts split: UC Berkeley’s Fyodor Urnov compliments disclosure; Stanford’s Le Cong notes PR ahead of queued experiments. Gladstone’s Seth Shipman credits speed of finding, not autonomy. Texas A&M’s Jason Gill previously identified the reverse transcriptase in 2021; novelty may be the repeats suggesting Crispr-like organization. Anthropic told the NYT Claude was not trained on user transcripts after a researcher’s suspicion.

We’re excited carefully.

Pattern search across huge genomic databases is exactly where language models can earn their keep — months of eyeball mining compressed into a day. That is useful AI for science. Calling it a Crispr-like discovery before function is proven is how you get sand that looks like diamonds on Santa Monica Beach, as Cong put it.

I’m watching the wet lab, not the X thread. Gene-editing utility is an empirical claim. Prior identification of the enzyme tempers the “Claude found it” mythology; scientists in the loop matter. Amodei’s musing about Claude eventually running lab equipment is a longer-horizon safety story of its own.

We’ll celebrate acceleration of hypothesis generation. We will not confuse a shiny sequence with a therapy. Biology’s clock — decades from Crispr sequences to one approved drug — still applies.

WIRED, September 29, 2026. Anthropic announcement September 23; expert quotes throughout. Retrons vs Crispr distinction noted by sources.