XNews

Claude Discovers CRISPR-Like ART Enzyme System

Anthropic announced that Claude discovered ART, a CRISPR-like enzyme system in bacteriophage DNA. About 950 Claude agents screened over 200,000 reverse transcriptases in 21 hours, and the human life sciences team ran follow-up experiments in its BSL-1/BSL-2 lab to validate the find.

Key insights

Claude Finds CRISPR-Like ART System in Phage DNA

Anthropic announced that Claude discovered a previously unnoticed enzyme system in bacteriophage DNA, named ART for array-associated reverse transcriptase. It comprises a reverse transcriptase, a partner gene of unknown function, and a long series of evenly spaced DNA repeats. Like CRISPR's repeat arrays that store targeting information, ART's array is transcribed into a set of distinct short RNAs, suggesting a potentially programmable mechanism. Systems combining these features have rarely been found before, and all known ones perform DNA cutting, copying, or pasting operations, some already developed into tools or gene-editing drugs. Anthropic acknowledged ART's precise function remains unclear.

Discovery Ran on Roughly 950 Claude Agents for 21 Hours

Researchers gave a single initial prompt asking Claude to find interesting new reverse transcriptases in massive DNA databases. About 950 Claude agents ran 21 hours and consumed 210 million tokens, screening over 200,000 reverse transcriptases into 3,500 candidates and then a shortlist of 20, written up in human-readable reports. One agent recognized the repeat array while reading raw sequences, counted repeats, measured spacing, compared against known systems, and checked the literature before submitting for human review. The reverse transcriptase itself had been reported before, but the adjacent array and partner protein are Claude's first findings.

Expert Endorsement and Compressed Timelines

Genome-mining work of this kind typically takes experts weeks to months; the agent swarm produced shortlisted candidates and reports in a single day. CRISPR pioneer and MIT professor Feng Zhang reviewed the preprint and called it a good example of AI agents participating in biological discovery, saying such arrays merit deeper study. Anthropic is soliciting research topics from external scientists for collaboration and stated other laboratories can in principle replicate the AI-proposes, human-validates workflow using publicly released tools.

Amodei Points to an Exponential Trend in AI Science

Dario Amodei wrote the work was mostly, though not entirely, done by Claude: the life sciences team suggested a broad research area, Claude read the literature and genome data, discovered the system, and proposed verification experiments the team carried out. He traced a pattern of AI going from weak to superhuman in new intellectual domains within years, citing math performance rising from high-school level in 2023 to national competition level in 2024, minor open problems in 2025, more significant ones in early 2026, and the top open problems in mathematics by late 2026, arguing AI for biology is on a similar exponential. He contrasted math's purely theoretical nature with biology's experimental requirement, saying human-AI collaboration handles validation in weeks, and that Claude may eventually safely control lab equipment with safeguards, though not today.

Discovery Aims at Accelerating the Biomedical Pipeline

Amodei framed fundamental biological discovery as the first stage of a pipeline spanning translational research, drug discovery, clinical trials, and medicine delivery, arguing that accelerating the first stage speeds and broadens the entire pipeline by identifying new drug targets, finding new therapeutic modalities, enabling more precise measurement, and shortening the experimental loop. Faster discovery increases candidate throughput even if clinical trial latency stays unchanged. He connected this to his Machines of Loving Grace goal of curing most diseases in 5-10 years, calling it barely possible if AI is applied at every pipeline stage. He placed ART in a lineage including CRISPR, bridge recombinases, and VIPR, noting a Stanford team independently described a similar but distinct reverse transcriptase system with a non-coding array.

Anthropic Runs Its Own BSL-1/BSL-2 Wet Lab

The work came from Anthropic's life sciences team formed this spring, which built a wet lab in the Bay Area restricted to BSL-1 and BSL-2 low-risk research, handling no pathogens infectious to humans, with all experiments performed by human scientists. The team's daily tools are the publicly released Claude Science and Claude Code. Amodei said Anthropic aims to grow the life sciences team and extend the approach to a broad range of problems, inviting research collaboration proposals.

Source posts