Claude Discovers Novel CRISPR-like Enzyme System
Anthropic announced that Claude discovered a new enzyme system in bacteriophage DNA, named ART for Array-Associated Reverse Transcriptase. About 950 Claude agents ran 21 hours on 210 million tokens, narrowing over 200,000 reverse transcriptases to 20 candidates for human validation.
Key insights
ART System Mirrors CRISPR Architecture in Bacteriophage DNA
Claude identified a class of enzyme systems in bacteriophage DNA that had not been previously noted, named ART for Array-Associated Reverse Transcriptase. The system comprises a reverse transcriptase, an adjacent partner gene of unknown function, and a long array of evenly spaced repeat DNA sequences. Experiments confirmed the array is transcribed into a set of distinct short RNAs, hinting at a programmable mechanism comparable to CRISPR's repeat arrays, which store navigation information for cutting the genome at targeted positions. The discovery was made by Claude reading raw sequences and confirmed through experiments carried out by human scientists.
Rare Feature Combination Points Toward a Possible Gene Editing Mechanism
Systems combining these features have rarely been found before, and all previously known ones perform DNA cutting, copying, and pasting operations; CRISPR has already produced gene editing drugs and several others are in development as tools. Anthropic describes ART as a molecular machine suspected to represent a new gene editing mechanism, while stating that its precise function and biotechnological utility are not yet clear. Dario Amodei characterized it as work he would have been proud to do as a PhD student and part of a line of research he believes is just beginning.
Discovery Ran Almost Entirely by Claude Across 950 Agents
The discovery process was nearly fully autonomous. Researchers gave only an initial prompt to search massive DNA databases for interesting new reverse transcriptases. About 950 Claude agents ran for 21 hours and consumed 210 million tokens, filtering more than 200,000 reverse transcriptases down to 3,500 candidates and then the 20 most worthwhile, written up in human-readable reports. One agent recognized the repeat array while reading raw sequences, counted the repeats, measured spacing, compared against known systems, and checked literature before submitting for human review. The reverse transcriptase itself had been reported before, but the adjacent array and partner protein are first identified by Claude. Such genome mining normally takes experts weeks to months.
Feng Zhang Endorses; ART Joins a Lineage of RT-Based Systems
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. Amodei's announcement places ART in a decades-long lineage beginning with CRISPR and continuing through the bridge recombinase and VIPR in recent years, amid heightened interest in reverse transcriptase-based systems. He noted that a Stanford team working independently recently described a novel RT system with an associated non-coding array similar in some ways to the one Claude found, though the two systems evolved independently, and stated the field is at the very beginning of finding such systems and developing them into biotechnology tools.
Amodei Argues AI in Biology Follows the Same Exponential as Math
Amodei writes that it is easy to dismiss the discovery as a one-off, but AI performance in new intellectual domains goes from weak to superhuman within a few years, citing the math trajectory from high-school level in 2023 to solving the top few open problems in mathematics by late 2026. He states AI for biology is on a similar exponential trend, and that the experimental requirement is addressed through human-AI collaboration validating key results in weeks, with Claude possibly eventually controlling lab equipment autonomously under safeguards. Accelerating fundamental discovery increases promising candidates entering the pipeline even if clinical trial latency is unchanged, a step toward the Machines of Loving Grace goal of curing most diseases in 5 to 10 years.
Wet Lab Setup Keeps Workflow Replicable With Public Tools
The work comes from Anthropic's life sciences research group formed this spring, which built its own wet lab in the Bay Area conducting only BSL-1 and BSL-2 level low-risk research, handling no pathogens capable of infecting humans, with all experiments performed by human scientists. The team's daily tools are the publicly released Claude Science and Claude Code, meaning other laboratories can in principle replicate the workflow of AI generating hypotheses in bulk with humans validating experiments. Anthropic is soliciting research topics for collaboration from external scientists and welcoming proposals to join the life sciences team.