Claude Discovers Novel CRISPR-Like Enzyme System in Phage Genomes: Anthropic Launches Wet Lab with "ART" Discovery

The Three-Question Framework

What Happened?

Anthropic announced the establishment of a dedicated life sciences research division and wet laboratory in the San Francisco Bay Area, sharing its first breakthrough: Claude agents autonomously discovered a previously uncharacterized biological enzyme system named Array-associated Reverse Transcriptases (ART). Tasked with exploring a massive database of over 200,000 bacterial and viral reverse transcriptases, approximately 950 parallel Claude agent instances spent 21 hours consuming 210 million tokens to analyze genomic neighborhoods, discovering a distinct tandem array of non-coding DNA repeats paired with an uncharacterized accessory protein in jumbo bacteriophages.

Why It Matters

For decades, programmable genetic tools—from restriction endonucleases and Taq polymerase to CRISPR-Cas9—were discovered through serendipitous observation of bacterial immune machinery. Claude's discovery of the ART system represents the first time foundation models have driven the autonomous hypothesis, genomic neighborhood mapping, and selection of a novel candidate enzyme system from raw sequencing reads, subsequently validated through wet-lab expression. CRISPR pioneer Feng Zhang (MIT & Broad Institute) confirmed the significance of the findings, noting that RNA-repeat arrays coupled to reverse transcriptases possess defining characteristics shared by programmable nucleic acid editing engines.

What Could Happen Next

Anthropic's wet-lab team is expressing the ART enzyme complex in standard bacterial hosts to determine its precise enzymatic mechanism—specifically whether the expressed short non-coding RNAs guide targeted RNA-to-DNA copying or viral defense—while expanding Claude's automated hypothesis-generation loop across uncharted protein families.

Executive Key Takeaways

Autonomous AI Biological Discovery: 950 parallel Claude agents combed 200,000 reverse transcriptases to isolate the novel ART enzyme system.

CRISPR-Like Structural Architecture: Identifies a tandem non-coding DNA repeat array and an accessory protein coupled to a viral reverse transcriptase.

Bay Area Wet-Lab Validation: Anthropic establishes dedicated BSL-1/BSL-2 molecular biology facility to experimentally verify AI candidate hypotheses.

Praise from CRISPR Pioneers: MIT and Broad Institute researcher Feng Zhang validates the significance of the AI-discovered RNA-repeat arrays.

On September 23, 2026, Anthropic officially announced the formation of a dedicated life sciences division and experimental wet laboratory in the San Francisco Bay Area. Accompanying the lab's public debut was a landmark announcement: Claude foundation models had autonomously discovered a previously uncharacterized family of biological enzymes named Array-associated Reverse Transcriptases (ART), exhibiting structural characteristics remarkably reminiscent of CRISPR gene-editing machinery.

Many of modern biotechnology's foundational tools originated from unusual microbial defense mechanisms. Restriction enzymes, which cut double-stranded DNA at exact recognition motifs, were isolated from bacterial immune barriers against viral bacteriophages. Taq polymerase, which functions at near-boiling temperatures to enable the Polymerase Chain Reaction (PCR), was found in hot-spring bacteria. And CRISPR itself was initially identified as an odd series of clustered repetitive DNA repeats before being harnessed into the Nobel Prize-winning genome editing revolution.

To test whether frontier artificial intelligence could systematize and accelerate biological discovery, Anthropic assembled a unified team of computational biologists and experimentalists. They orchestrated an automated research harness comprising approximately 950 Claude agent instances running in parallel. The agents were tasked with an open-ended mandate: survey an uncurated database containing more than 200,000 microbial and viral reverse transcriptases (RTs)—enzymes capable of transcribing RNA back into DNA—and surface functionally anomalous candidates.

Consuming 210 million tokens over 21 hours of autonomous evaluation, the multi-agent system whittled 200,000 sequences down to 3,500 candidates, and finally to 20 top-priority structural anomalies. While inspecting the genomic neighborhood of a jumbo bacteriophage RT, an agent flagged a striking architectural anomaly: '[The DNA next to the RT] is spectacular: I can see by eye a tandem repeat array … that's a CRISPR-like … repeat array?!'

Rather than hallucinating speculative biology, the agent systematically tallied the repeats, measured inter-spacer distances, cross-referenced all known biochemical literature, and proved that this non-coding array—flanked by an uncharacterized partner accessory protein—constituted an entirely undescribed functional operon. When Anthropic's laboratory scientists expressed the sequence in vitro, RNA sequencing confirmed that the array is transcribed into distinct short RNA fragments, strongly suggesting an active, programmable targeting mechanism analogous to CRISPR guide RNAs.

CRISPR pioneer Professor Feng Zhang of MIT and the Broad Institute publicly commended the research: 'This is an exciting example of how AI agents can contribute to biological discovery. The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigation.' Anthropic's validation marks a permanent inflection point where autonomous foundation models transition from passive literature assistants into active, hypothesis-generating experimental discovery engines.

The Three-Question Framework

What Happened?

Why It Matters

What Could Happen Next

Executive Key Takeaways

Primary Sources & Citations