Claude found a CRISPR-like enzyme system — and can't say what it does

Share
Claude found a CRISPR-like enzyme system — and can't say what it does

Anthropic's lab published an unfinished result and named it anyway, while Google turned voice cloning into a line item you can price by the hour.

Anthropic says one of its Claude agents autonomously spotted a previously uncharacterized enzyme system buried in a DNA database — and the company published the finding before it knows what the system does. The search took roughly 950 Claude agents 21 hours and 210 million tokens: they gathered more than 200,000 reverse transcriptases, picked out 3,500 candidate systems, and narrowed those to the 20 most compelling. Then one agent, reading raw DNA beside an odd-looking reverse transcriptase, reported the find in its own words: "The DNA next to the RT is spectacular: I can see by eye a tandem repeat array … that's a CRISPR-like … repeat array?!" Lab work confirmed a system in bacteriophages with three parts — the reverse transcriptase, a partner gene, and a long array of evenly spaced repeats — which Anthropic calls array-associated reverse transcriptases, or ART. Early experiments show the array is expressed as a set of distinct short RNAs, the same trick CRISPR uses to make itself programmable.

Anthropic's own framing is notably careful: the primary function of ARTs is still unknown, experiments are ongoing, and the company says it shared early "to demonstrate Claude's capabilities." The CRISPR comparison is Anthropic's, not the field's. Feng Zhang, the MIT and Broad Institute professor who helped pioneer CRISPR editing, called the work "genuinely intriguing and merits further investigation" and said he hoped it would encourage other scientists — an endorsement that is measured, not a confirmation. All of the wet-lab work was done by human scientists at a Bay Area lab working at the two lowest biosafety levels, with no human pathogens involved. Read alongside the timing — Anthropic is preparing to go public — the preprint reads as a capability demo as much as a biology result. We covered the lab itself when it surfaced this month — Anthropic quietly built a wet lab for its AI drug push.


Google shipped two speech models, Gemini 3.8 Flash TTS and Flash-Lite TTS, that replace preset voices with voices you describe in a sentence. Flash TTS designs a voice from a text prompt across more than 100 languages and dialects, offers a library of 2,000-plus presets including regional variants like Mexican Spanish and Quebec French, and can clone a speaker from a 30-second sample — but only after that person records a spoken consent statement that the system checks against the sample. Every generated clip carries an inaudible SynthID watermark, and replicated voices also carry C2PA content credentials. Google claims first place on Hume AI's Voice Design Benchmark with a score of 71.4.

The pricing is the part that travels: an hour of generated audio costs about $0.81 on Flash TTS and $0.54 on Flash-Lite through the end of 2026, then doubles on January 1 — a rate that puts real pressure on the specialist voice vendors. Google is rolling both models out through the Gemini API and AI Studio, with enterprise access to follow, and voice replication is geo-blocked in several jurisdictions including the UK, the EEA, Illinois and Texas. We covered the sibling launch that set the price benchmark this month — Gemini 3.8 Live tops voice benchmarks at a sixth of GPT-Live's price.

What to watch: whether ART turns out to do anything programmable — and whether Anthropic's next science announcement arrives with a function attached.

If an agent flags a discovery that humans then have to characterize, whose discovery is it? Tell us in the comments.

Sources: Anthropic — Claude discovers a novel enzyme system with CRISPR-like repeats · Anthropic technical report (PDF) · The Verge · Google — Gemini 3.8 text-to-speech · The Next Web · The Decoder