Feng Zhang 1 item
‹ All posts People /Feng Zhang /News ARCHIVE
News · Wed, 23 Sept 2026 · 15:30 ET

Claude finds phage ART enzyme with CRISPR-like repeats

Anthropic primary: Claude agents spotted array-associated reverse transcriptases in jumbo phage DNA; lab verified; function still open.

primary Dario AmodeiFeng Zhang AnthropicClaudeCRISPR Source ↗
Outlet: AnthropicPublished: Wed, 23 Sept 2026 · 00:00 ET

Opening

Anthropic says Claude autonomously found a novel bacteriophage enzyme system with CRISPR-like DNA repeats — function unknown; Feng Zhang calls the RNA-repeat / reverse-transcriptase pairing “genuinely intriguing.” Anthropic’s Sep 23 post introduces a Bay Area life-sciences lab (BSL-1/BSL-2; humans run wet lab) and reports early results: Claude agents, given a high-level prompt to mine reverse transcriptases (RTs), flagged a jumbo-phage RT with an adjacent evenly spaced DNA-repeat array and an accessory protein — a pattern Anthropic names array-associated reverse transcriptases (ART). The underlying RT had appeared in prior studies; Anthropic says Claude first noticed the associated non-coding repeat array. After lab expression work, Anthropic says the array is expressed as distinct short RNAs. Dario Amodei’s X post stresses utility and significance are not yet clear, and frames biology as on a similar multi-year “weak → superhuman” curve as math. The post quotes Feng Zhang (MIT / Broad) reviewing a pre-print.

The numbers

Search campaign — ~950 agents, ~210 million tokens, ~21 hours over a massive RT DNA database — Anthropic primary.

Candidate funnel — >200,000 RTs gathered → ~3,500 new candidate systems → 20 most-compelling human-readable reports — Anthropic primary.

Wet lab — human scientists only at BSL-1/BSL-2; no autonomous equipment control claimed today — Anthropic primary.

Why it matters

This is a capability / go-to-market signal for Anthropic’s life-sciences push ahead of the expected IPO window already in the archive: an end-to-end claim that agents can mine genomes, propose candidates, and hand humans a wet-lab verification loop in weeks. It does not establish a programmable gene-editing tool — Anthropic and Zhang both mark function as open. Self-serving framing is the product demo itself; the Zhang quote is third-party but invited.

Connects to

Anthropic ~$100B ARR / Nov IPO path — life-sciences lab launch and public discovery demo sit beside the IPO narrative already documented (news).

Cheaper Opus 5.5 list — same-week model/pricing moves; separate from this biology claim (news).

The Open/Close  ·  Research commentary, not investment advice. Positions may be held in securities mentioned.