Anthropic's AI, Claude, discovers a new enzyme system in bacteriophages, named ART, featuring CRISPR-like DNA structures.
Anthropic's AI, Claude, has discovered a new enzyme system within the DNA of bacteriophages, the viruses that infect bacteria. This system, designated as array-associated reverse transcriptases (ART), features repetitive DNA sequences reminiscent of CRISPR.
Claude has uncovered a previously unknown enzyme system in the DNA of bacteriophages. The enzyme’s gene is accompanied by a long sequence of repeating DNA—a structure similar to CRISPR.
We still don't fully understand the function of this system, but…
— Anthropic (@AnthropicAI) September 23, 2026
Discovery Process of ART
Claude was tasked with identifying significant examples of reverse transcriptases—enzymes that convert RNA into DNA—by autonomously analyzing a vast genomic dataset of approximately 1.9 billion protein clusters. This endeavor involved around 950 AI agents working for 21 hours, leading to the compilation of over 200,000 reverse transcriptases and narrowing it down to 20 promising candidates for further research.
Anthropic estimates that a human researcher would require weeks or even months to complete a similar analysis. During the investigation of one candidate, Claude identified repeating DNA sequences near an unusual reverse transcriptase gene. After measuring intervals between these repeats and comparing them to known structures, Claude validated the findings with existing scientific literature before presenting them to researchers.
This reverse transcriptase had been previously noted in studies of a giant bacteriophage, but the significant advancement lies in identifying a larger system that includes this enzyme, an unknown adjacent gene, and a series of uniformly spaced repeating DNA segments.
All experiments were conducted by Anthropic's lab, which operates under BSL-1 and BSL-2 biosafety conditions and does not engage with pathogens harmful to humans.
CRISPR Connections
CRISPR technology was initially recognized through unusual repetitive sequences in microbial genomes, which were later found to help confer immunity against reinfection from bacteriophages. This mechanism was subsequently adapted for genome editing applications.
For the newly discovered ART, researchers also identified a long array of repeating sequences, with early experiments indicating the production of individual short RNAs. This similarity to CRISPR has led to comparisons, although the precise function of ART remains unclear. Anthropic has not yet demonstrated any capabilities for DNA editing or sequence recognition associated with this system.
Anthropic's Commitment to Life Sciences
The identification of ART aligns with Anthropic's growing focus on life sciences. On September 17, the company launched a public beta for its Life Sciences Verification Program, allowing research teams to access models like Mythos 5.1 and Opus 5 with fewer restrictions for biological applications.
Today we’re opening applications for the Life Sciences Verification Program.
This initiative enables life science professionals to utilize our models with new safeguards for a comprehensive range of biology-related tasks.
— Anthropic (@AnthropicAI) September 17, 2026
The company also introduced a High-risk Use tier for projects with greater potential for misuse. Furthermore, Claude has optimized over 30 open-source models for biomolecular research, enhancing their performance significantly within a month.
Anthropic is also collaborating with Adaptyv Bio to host a protein design competition, aiming to validate over 5,000 proposed structures, with substantial funding provided for successful candidates. The company has formed a dedicated team to explore how AI can accelerate discoveries in biology, alongside establishing its own lab in the San Francisco Bay Area for hypothesis testing and research.
Earlier this year, the Claude Science environment was launched to support researchers, with plans to develop therapies for advanced diseases. Notably, Stanford University researchers recently utilized AI to design complete bacteriophage genomes, with promising results against E. coli.
