Anthropic says Claude found a phage enzyme system nobody had put together

Anthropic published a claim on Wednesday that will be argued about for a while: Claude, running as roughly 950 agents over 21 hours and consuming 210 million tokens, worked through more than 200,000 reverse transcriptases, narrowed them to about 3,500 candidate systems, then to twenty finalists, and identified a bacteriophage enzyme system it calls array-associated reverse transcriptases, or ARTs.

The novelty claim is narrower than the headlines suggest, and Anthropic says so itself. The underlying reverse transcriptase had already been identified in previous studies of a jumbo phage. What Anthropic claims is new is that Claude appears to be the first to notice the system's defining features together — an associated array of non-coding DNA and an accessory protein of unknown function. “Although we don't yet know its function,” the company writes, “the system that Claude discovered has a set of characteristics that have only ever been found together in a handful of other systems, all of which are programmable and perform operations like cutting, copying, and pasting DNA.” That is the CRISPR-shaped implication, stated as an implication.

Feng Zhang of MIT and the Broad Institute, one of the people who made CRISPR a tool, reviewed the findings and called the identification of RNA-repeat arrays associated with reverse transcriptases “genuinely intriguing” and said it “merits further investigation”. That is encouragement, not validation, and he reviewed rather than co-authored.

Two caveats belong in any honest write-up. The paper is a preprint, and not on bioRxiv or any preprint server — it is a PDF hosted on Anthropic's own CDN. And the function is undetermined; Anthropic says its work to understand what ARTs actually do is ongoing.

• Confirmed About 950 agents consumed 210 million tokens over 21 hours, gathered 200,000+ reverse transcriptases, narrowed to 3,500 candidate systems and then 20 finalists. Anthropic
• Confirmed The underlying reverse transcriptase was already known from prior studies; the new claim is that Claude appears to be the first to notice the system's defining features together. Anthropic
• Claimed The work is a preprint hosted on Anthropic's own CDN rather than a preprint server, is not peer-reviewed, and the system's function is undetermined. Anthropic preprint