Claude Discovers New Enzyme System in Bacteria Viruses

Anthropic announced that its AI model Claude identified a previously unknown enzyme system within the DNA of viruses infecting bacteria. Previous research had documented the core of this enzyme but overlooked surrounding elements.
The company revealed these findings on September 23. This discovery serves as significant evidence that AI can now detect patterns overlooked by human experts. But what does this actually mean for real-world applications?
Many medical breakthroughs began when scientists noticed anomalies in nature. For instance, enzymes extracted from Yellowstone hot springs became the foundation of PCR technology, which underpins most modern medical testing.
Claude followed general instructions from Anthropic scientists to identify such patterns. CEO Dario Amodei described this as part of a broader trend: AI models struggled with high school-level math in 2023 but will begin solving the toughest unsolved problems in their fields by late 2026. He believes AI in biology is following a similar growth curve.
If Amodei is correct, faster discoveries could reveal new drug targets and treatment forms. While clinical trial timelines won't shorten, more promising candidates may enter the drug development pipeline.
Amodei previously stated that AI could contribute to treating most diseases within 5-10 years, calling this goal achievable.
Currently, all experiments are conducted by human scientists, and the exact function of this new system remains unknown.
What is CRISPR?
Bacteria face constant attacks from viruses called bacteriophages. Many bacteria defend themselves using a system known as CRISPR.
CRISPR functions like a memory storage system. Bacteria store short DNA fragments from past invaders between repeated sequences, which scientists call an 'array.'
Each fragment is copied into short RNA that guides proteins to cut matching viral DNA if the same virus re-infects.
Scientists first discovered CRISPR in bacteria as a unique pattern of repeating DNA. Later, they learned to manipulate the system using custom RNA guides, turning it into a programmable gene-editing tool used widely in medicine today.
What did Claude discover?
Claude's discovery centers around an enzyme called reverse transcriptase, which copies RNA back into DNA. Bacteria use various reverse transcriptases to fight viruses, according to Anthropic.
Researchers had already documented this enzyme in 'jumbo phages,' unusually large viruses. However, previous studies missed key features of the system. Claude appears to be the first to identify it.
The newly identified system includes an uncharacterized companion protein and a specific pattern of repeated DNA sequences (long array) arranged similarly to CRISPR arrays. While CRISPR is mainly found in bacteria, ART is primarily discovered in phages that infect bacteria.
Anthropic named this three-component system 'array-associated reverse transcriptases,' abbreviated as ART. When the Claude agent made this discovery, it recorded its surprise while reading raw DNA data.
How did Claude discover it?
The AI agent was one of approximately 950 agents working on the task. Anthropic scientists instructed the AI to search a large DNA database for new reverse transcriptases using a single prompt.
Over 21 hours, the agents processed over 210 million tokens (small text fragments handled by AI models). This resulted in collecting more than 200,000 reverse transcriptases and selecting 3,500 promising new systems. The list was narrowed to 20 top candidates for human review.
Anthropic states that this analysis would typically take scientists weeks or months. The agent worked similarly to a human scientist: counting repeats, measuring intervals, comparing layouts with existing systems, and searching literature for prior reports.
The experiments were conducted at Anthropic's Bay Area research facility, which does not handle pathogens that infect humans.
What have they discovered so far?
The initial experiment provided early clues about how ART might function. Researchers confirmed that ART's repeat arrays convert into sets of short RNAs with specific targeting roles—similar to CRISPR.
Anthropic suggests this implies ART may operate similarly, though the system's main function remains unknown as further experiments are ongoing.
The combination of features in ART is extremely rare. Anthropic states that very few such systems are known to exist.
All these systems can be programmed to cut, copy, or paste DNA—similar to CRISPR and other emerging tools. However, Anthropic has not yet demonstrated that ART performs these functions.
What does the CEO think?
CEO Dario Amodei stated on X that Anthropic suspects ART could be a new gene-editing mechanism but acknowledged its function, usefulness, and importance remain unclear.
He highlighted recent work by Stanford researchers who found similar reverse transcriptase systems with non-coding arrays. While these resemble ART in some ways, the two systems evolved independently.
Is this the next CRISPR?
Peng Zhang, a pioneer of CRISPR at MIT and Broad Institute, reviewed an unpeer-reviewed preliminary paper on the findings. He found the RNA repeat arrays interesting but noted more research is needed.
Kevin Blake, a microbiologist at the University of Washington School of Medicine, expressed more skepticism. He stated that natural CRISPR differs significantly from technological CRISPR and emphasized that millions of bacterial species remain unstudied, leaving many CRISPR-like sequences unclassified.
Anthropic says further experiments are underway to determine ART's function—whether it represents a new programmable tool or is simply a unique feature in phage biology.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 클로드, 새로운 효소 시스템 발견…기능 미지수