Claude Discovered a New CRISPR-Like System: What ART Is and What We Still Don't Know

By Carlos Montiel | Enterprise AI Specialist
Leer en español →
Published: 2026-09-24 | By: Carlos Montiel | Reading time: ~6 min

Anthropic announced that Claude, guided by researchers at its San Francisco biology lab, autonomously identified a new enzyme system in bacteriophage DNA that could represent a previously undescribed gene-editing mechanism, with a repeat architecture similar to CRISPR. Anthropic itself is clear on one thing: it still doesn't know what the system actually does.

What Claude actually found

The system was named ART (array-associated reverse transcriptases). It's made up of three parts: a reverse transcriptase enzyme (which copies RNA into DNA), an associated gene of still-unknown function, and a series of DNA repeats at regular intervals — the same "repeats + enzyme" architecture that characterizes CRISPR, though that doesn't mean ART works the same way. According to Anthropic, the system appears mostly in the genomes of bacteriophages (viruses that infect bacteria).

How it was found: not a hunch, brute force at scale

Discovery numbers (per Anthropic): - ~950 Claude agents working in parallel - 21 hours of searching genomic databases - ~210 million tokens used - Over 200,000 reverse transcriptases collected - ~3,500 candidate systems identified before isolating ART

The process was essentially a massive, systematic search: hundreds of agents reviewing and classifying hundreds of thousands of genomic sequences in parallel — the kind of work that would take a human bioinformatics team far, far longer. It was the scale of the search, not a single flash of intuition, that made it possible to isolate one interesting candidate out of thousands of possibilities.

What Anthropic is NOT saying

Anthropic is explicit: it has not determined ART's actual function. We know today that it's "CRISPR-like" in architecture (repeats + associated enzyme), but there's no evidence yet that it's "CRISPR-like" in function — meaning it hasn't been confirmed that it can edit genes in a targeted way the way CRISPR-Cas9 does. The data also hasn't been peer-reviewed. It's a promising candidate for future research, not a ready-to-use gene-editing tool.

Why this matters beyond biology

The real value of this news for any company evaluating AI isn't genetic, it's operational: it's another concrete example of a pattern we already covered with OpenAI's Navier-Stokes case — swarms of hundreds or thousands of AI agents doing systematic search at a scale no human team can match in the same timeframe, followed by a validation step (peer review, in this case) before any conclusion is accepted as true. The business lesson is the same: agent-scale search is already real and technically viable, but its value depends entirely on having an objective verification mechanism at the end of the process.
Carlos Montiel
Enterprise AI Solutions Architect
LLMs, Agents & Orchestration Specialist
guatemalia.com/#contacto · info@guatemalia.com

Need to implement AI in your company?

Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.

Contact Carlos Montiel

info@guatemalia.com