Anthropic has used hundreds of AI agents to search billions of biological sequences, uncovering unusual DNA patterns in giant viruses that could eventually point to new molecular tools. Artificial intelligence is increasingly moving beyond software and data analysis and into the laboratory, with Anthropic now exploring how autonomous AI agents can help scientists identify previously overlooked biological discoveries.
The company’s new life sciences research group has reported an unusual finding involving DNA sequences in several giant viruses. The sequences resemble patterns associated with CRISPR immune systems found in some microorganisms, although researchers say it is far too early to conclude that the viral DNA performs the same function.
The discovery came from an experiment involving approximately 950 AI agents, which spent more than 21 hours searching through genomic data and examining billions of proteins.
The work was announced on September 23, 2026, alongside the launch of Anthropic’s biology-focused wet laboratory. The research has been posted as a preprint and has not yet undergone peer review.
AI agents search for biological “weirdness”
Anthropic’s approach is different from simply asking an AI model to analyse a dataset.
The company deployed autonomous AI agents capable of independently exploring biological information, discussing preliminary findings and determining what areas deserved additional investigation.
Researchers instructed the agents to examine DNA sequences associated with billions of proteins and search for proteins that could potentially work alongside enzymes known as reverse transcriptases.
During that process, the agents noticed a repeated DNA sequence near a reverse transcriptase gene in a giant virus.
Rather than treating the sequence as an isolated anomaly, the agents investigated further and identified similar patterns in the genomes of other viruses.
That repeated pattern is what attracted the researchers’ attention.
Eric Kauderer-Abrams, who leads Anthropic’s life sciences work, described the objective as creating a systematic way of scaling up the biological research process of looking for unusual or unexpected phenomena.
Why the CRISPR connection matters
CRISPR has become one of the most important technologies in modern molecular biology.
Naturally occurring CRISPR systems help certain microorganisms defend themselves against genetic invaders. Scientists have subsequently adapted components of these systems into powerful gene-editing technologies capable of making targeted changes to DNA.
The sequences identified by Anthropic’s AI agents share some structural similarities with DNA arrangements found in certain CRISPR systems.
But there is an important distinction.
Researchers have not established that the viral sequences actually function like CRISPR.
The similarities currently represent a scientific lead rather than a confirmed biological mechanism.
In known CRISPR systems, repeated DNA sequences can be separated by genetic material originating from previous viral or other genetic invaders. RNA produced from those sequences can help guide enzymes towards invading DNA, allowing the system to target and cut it.
Anthropic’s newly identified viral sequences do not currently have enough evidence behind them to establish that they operate in the same way.
Researchers also have not identified a known DNA-cutting enzyme working alongside the sequences.
Discovery is only the beginning
The most significant part of the experiment may therefore not be the discovery itself, but what happens next.
Finding an unusual DNA pattern is relatively different from proving what that pattern does.
Scientists will need to isolate and study the relevant biological components, conduct laboratory experiments and determine whether the sequences have a useful function.
That process highlights an important limitation of AI-driven biology.
AI systems can search enormous datasets far faster than human researchers could manually examine every sequence. They can identify relationships and patterns that might otherwise remain hidden in vast biological databases.
But biological hypotheses still need to be tested in the physical world.
Anthropic’s Kauderer-Abrams has emphasised that distinction, noting that life-sciences research ultimately requires experiments to establish how biological systems actually behave.
From AI analysis to AI-assisted laboratories
Anthropic’s experiment represents part of a broader shift in the relationship between artificial intelligence and scientific research.
Large language models have already demonstrated an ability to process scientific literature, analyse information and assist researchers with complex computational tasks.
The next stage involves AI agents that can operate with greater autonomy, dividing research questions into smaller tasks, exploring datasets, comparing results and deciding which findings deserve further attention.
Anthropic’s new wet-lab initiative takes that concept into the physical world, bringing AI-assisted discovery together with human scientists and laboratory experiments.
The potential is significant.
Biological databases contain enormous quantities of information, much of which has not been fully characterised. AI systems capable of systematically searching those datasets could help researchers identify proteins, enzymes and genetic mechanisms that might eventually become useful in medicine, biotechnology or other areas of science.
But the discovery-to-application journey can be long.
A sequence that looks interesting computationally may turn out to have no useful biological function. Even when a function is confirmed, converting it into a practical biotechnology tool can require years of experimentation and development.
What happens next?
For Anthropic’s latest discovery, the immediate task is to determine what the repeated viral DNA actually does.
Researchers will need to establish whether the sequences are functional, identify any proteins or enzymes associated with them and determine whether they perform a role comparable to components of microbial CRISPR systems.
If laboratory experiments confirm a previously unknown biological mechanism, the finding could provide scientists with another naturally occurring molecular system to study.
That could eventually have implications for biotechnology, although any such applications remain speculative at this stage.
For now, the more immediate lesson is about the changing role of AI in scientific discovery.
Anthropic’s experiment demonstrates that AI agents can be deployed to explore biological information at a scale that would be extremely difficult for individual researchers to match manually.
The challenge is turning those computational discoveries into experimentally verified science.
AI may be getting better at finding the biological needle in the haystack. But scientists still have to pick it up, examine it and determine whether it is actually useful.
