Last Updated: 28th September, 2026
What Did Claude Discover? Anthropic says a group of Claude AI agents helped identify a previously uncharacterized biological system in bacteriophages, viruses that infect bacteria.
The system, which Anthropic calls array-associated reverse transcriptases (ART), combines a reverse transcriptase with a partner gene and a long array of repeating DNA sequences. The arrangement has some similarities to CRISPR-like systems, but ART has not been shown to be a new CRISPR system or a gene-editing technology.
The discovery came from an autonomous computational search in which Claude agents examined enormous amounts of biological sequence data. Anthropic says the campaign involved 949 agent sessions over about 21.5 hours and more than 215 million tokens, while searching across roughly 1.9 billion protein clusters. Human scientists subsequently performed the laboratory experiments.
The most important caveat is that scientists do not yet know exactly what ART does. Anthropic describes the result as an early biological discovery that requires further investigation.
What Happened?
On September 23, 2026, Anthropic announced an early result from its new life-sciences research program: Claude agents had helped identify a previously uncharacterized enzyme system associated with repeating DNA sequences.
The announcement is significant because the AI was not simply being used to summarize published scientific papers.
Instead, Claude agents were used to search genomic data, investigate candidate biological systems, identify unusual patterns, compare findings against existing research, and produce candidates for human scientists to investigate.
Anthropic says the project began with a broad research instruction to search DNA sequence databases for interesting examples of reverse transcriptases.
From there, the agents divided the work into many computational tasks and investigated potential candidates.
One agent eventually noticed something unusual in DNA located near a reverse-transcriptase gene: a repeating sequence pattern that resembled structures found in some biological systems associated with CRISPR.
That observation led researchers to investigate the system further.
What Is the New Enzyme System?
Anthropic calls the system array-associated reverse transcriptases, or ART.
It consists of three major components:
- A reverse transcriptase
- An associated partner gene
- A long array of repeating DNA sequences
A reverse transcriptase is an enzyme that can use RNA as a template to produce DNA.
The important part of the finding is not simply that Claude found a new reverse transcriptase. Anthropic says the underlying reverse transcriptase had already appeared in previous research.
The new observation was the broader combination of the reverse transcriptase, partner gene and unusual repeat array.
The technical preprint describes ART as a family of jumbo-phage reverse transcriptases associated with repeated DNA units and a dedicated partner gene.
Why Are the Repeats Important?
The DNA repeat array is what makes the system particularly interesting.
Anthropic reports that the arrays contain repeated DNA units separated by spacers. The arrangement is reminiscent of repeat structures found in CRISPR systems.
But similarity in structure does not mean ART is CRISPR.
At this stage, researchers do not know whether ART performs anything resembling the programmable DNA operations associated with CRISPR systems.
That distinction is important because headlines suggesting that Claude has already discovered a new gene-editing tool would go beyond what has been demonstrated so far.
How Did Claude Agents Find It?
The process is one of the most interesting parts of the story.
Anthropic gave Claude a high-level research objective and allowed its agents to perform much of the computational investigation.
According to Anthropic’s announcement and technical report, the system:
- Surveyed huge collections of protein and DNA sequence data.
- Investigated reverse-transcriptase families.
- Identified candidate systems and genomic relationships.
- Examined sequences surrounding interesting candidates.
- Detected an unusual repeating DNA pattern.
- Compared the pattern with known biological systems.
- Produced a candidate for human scientific investigation.
- Had human researchers conduct laboratory experiments.
Anthropic reports that Claude agents examined more than 200,000 reverse transcriptases during the search and narrowed the work to a much smaller group of promising candidates.
The technical report gives a broader picture of the search campaign, describing an autonomous agent workflow operating across approximately 1.9 billion protein clusters.
The Moment Claude Spotted the Unusual DNA Pattern
The discovery reportedly happened when one Claude agent inspected DNA near a reverse-transcriptase gene.
Instead of finding only the expected information about the gene, the agent noticed a repeated pattern in the surrounding sequence.
Anthropic says the agent recognized the sequence as a potential tandem repeat array and began investigating whether the arrangement had been described previously.
This is an important detail because it illustrates a different potential use for AI agents in scientific research.
The agent wasn’t necessarily given the instruction:
“Find a CRISPR-like system.”
Instead, it encountered an anomaly during a broader investigation and decided that the anomaly was worth following.
That ability to pursue unexpected patterns is one of the ideas Anthropic highlights in its description of the research.
What Did Human Scientists Do?
This is where the headline needs some important context.
Claude did not independently conduct the physical laboratory experiments.
Anthropic’s computational agents searched the biological data and generated hypotheses and candidates.
Human scientists then performed the laboratory work needed to investigate the candidate system.
Anthropic explicitly says that all of the laboratory work was performed by human scientists in its Bay Area molecular biology laboratory.
The company’s researchers investigated the system experimentally and reported evidence that the repeat array can be expressed as distinct short RNAs.
However, the primary biological function of ART remains unknown.
That means the story is better understood as AI-assisted scientific discovery and hypothesis generation, rather than an AI independently creating and validating a new biological technology.
Why the Claude Discovery Matters
The biggest implication may not be ART itself.
It may be the research workflow used to find it.
Scientific databases are becoming too large for researchers to manually inspect every potentially interesting sequence.
AI agents can potentially help researchers search these datasets, identify unusual relationships and prioritize candidates for experimental investigation.
Anthropic’s research demonstrates one possible model:
Large dataset → AI agents → anomaly detection → hypothesis → human verification → laboratory experiment
This approach could eventually be useful in fields such as genomics, protein research, drug discovery and molecular biology.
But those potential applications should be treated as possibilities rather than established outcomes from this particular discovery.
How This Is Different From Traditional AI Chat
A normal chatbot interaction might look like this:
Researcher:
“Explain reverse transcriptases.”
AI:
Provides an explanation based on its available information.
The Claude research workflow was considerably different.
The agents were given tools and tasks that allowed them to:
- Search large datasets
- Retrieve biological sequences
- Analyze candidate genes
- Compare findings
- Delegate tasks
- Review other agents’ work
- Generate reports
- Pursue unusual observations
That makes the experiment an example of agentic AI applied to scientific research.
Instead of producing one answer, the AI system was used as part of a longer research process.
For readers learning about prompting and AI agents, this distinction is important.
You can learn more about how AI systems follow structured instructions in Promptenza’s guide to Prompt Engineering, and explore the components that make instructions clearer in Prompt Anatomy.
What Is a Reverse Transcriptase?
A reverse transcriptase is an enzyme that produces DNA using RNA as a template.
Reverse transcriptases occur in many biological systems and have been studied extensively.
In this case, the interesting question wasn’t simply whether a reverse transcriptase existed.
Researchers were interested in what other genetic components appeared alongside it.
The Claude agents’ discovery centered on the unusual organization around the reverse transcriptase and the associated repeat array.
That distinction helps explain why the research could identify something new even though the underlying enzyme family had already been studied.
Is ART a New CRISPR System?
Not based on the evidence currently available.
ART has features that remind researchers of some CRISPR-related systems, particularly the presence of repeated DNA sequences.
Anthropic itself describes the system as having properties reminiscent of CRISPR, rather than identifying it as CRISPR.
The company also says that the function of ART is still unknown.
So it would be premature to describe ART as:
- A replacement for CRISPR
- A proven gene-editing technology
- A new gene-editing tool
- A technology that can treat disease
- A commercially usable biotechnology
Those possibilities would require substantially more experimental evidence.
Could ART Become a Gene-Editing Tool?
That is an open question.
Some biological systems with unusual combinations of enzymes and nucleic-acid structures have eventually proved useful for biotechnology.
Anthropic points to the history of systems such as CRISPR as an example of how unusual biological mechanisms can eventually become useful technologies.
But there is currently no evidence that ART can be used as a gene-editing technology.
Researchers first need to determine what the system actually does.
That could involve understanding:
- What the reverse transcriptase does within the system
- What role the repeat array plays
- What the partner protein does
- How the associated RNAs function
- What biological process the system participates in
- Whether the system has any programmable activity
Those are research questions, not established capabilities.
What We Know vs. What We Don’t Know
| What is currently reported | What remains unknown |
|---|---|
| Claude agents identified an unusual biological system | The system’s precise biological function |
| Anthropic calls it array-associated reverse transcriptases (ART) | Whether ART has gene-editing capabilities |
| ART is associated with reverse transcriptases | Exactly how its components interact |
| The system contains a repeating DNA array | Why the repeat array exists |
| The system occurs in bacteriophages | Its potential biotechnology applications |
| Human scientists performed laboratory experiments | Whether ART could become a practical tool |
| The array appears to produce distinct short RNAs | What those RNAs actually do |
This distinction is particularly important when interpreting AI-generated scientific discoveries.
A candidate discovery is not the same thing as a proven technology.
What Makes This AI Discovery Different?
One of the most interesting aspects of the project is the scale of the search.
Anthropic says roughly 949 agent sessions worked through the research campaign for approximately 21.5 hours, consuming more than 215 million tokens.
That illustrates a potential advantage of agentic AI:
Parallel investigation
Instead of one researcher manually checking candidates one at a time, many AI sessions can investigate different pieces of a large dataset simultaneously.
Persistent research workflows
Agents can perform multiple steps rather than simply generating one response.
Anomaly detection
The system can potentially identify relationships that weren’t explicitly specified in the original task.
Human verification
Scientists can then concentrate their time on the candidates that appear most interesting.
This is similar to the broader shift from AI as a chatbot toward AI as a research assistant capable of carrying out multi-step workflows.
For a deeper explanation of how these workflows are changing AI use, see Promptenza’s AI Prompting Techniques.
The Biggest Limitation: AI Still Needs Scientific Verification
The discovery also demonstrates why human oversight remains important.
AI agents can process huge quantities of information, identify patterns and generate hypotheses.
But identifying an interesting pattern does not automatically establish its biological meaning.
Researchers still need experiments to determine whether the hypothesis is correct.
Anthropic’s own description makes this distinction clear: the company’s scientists performed the laboratory work and are continuing experiments to understand ART.
This creates an important model for AI-assisted science:
AI searches and proposes → humans evaluate → experiments test → evidence determines what survives.
What This Means for the Future of AI Agents
The Claude enzyme discovery points toward a broader question:
What happens when AI agents can search, reason, experiment computationally and coordinate complex research tasks?
Today’s AI systems are increasingly being used for more than writing and question answering.
Agents can already be designed to:
- Search information
- Analyze large datasets
- Write and execute code
- Use external tools
- Coordinate multiple subtasks
- Evaluate intermediate results
- Produce structured research reports
The Anthropic biology project takes that concept into scientific research.
The important development isn’t simply that an AI model recognized a DNA pattern.
It is that an AI-agent workflow helped move from a massive dataset to a biological hypothesis that human scientists could investigate experimentally.
What Claude’s Enzyme Discovery Does—and Doesn’t—Prove
It demonstrates
- AI agents can search extremely large biological datasets.
- AI can help identify unusual patterns in genomic data.
- Agents can generate candidates for experimental research.
- AI can contribute to multi-stage scientific workflows.
- Human scientists can use AI-generated hypotheses to prioritize research.
It does not yet demonstrate
- That ART is a gene-editing technology.
- That ART can replace CRISPR.
- That Claude independently conducted wet-lab science.
- That ART has a known therapeutic application.
- That the system has a confirmed practical biotechnology use.
The distinction matters because the scientific significance of ART will ultimately depend on experimental results.
What This Means for AI Prompting
There is also a useful lesson here for anyone learning how to work with AI.
The project wasn’t based on a simplistic instruction such as:
“Find something new in biology.”
The agents operated within a structured research workflow involving defined objectives, tools, searches, analysis and follow-up investigation.
That is closely related to the principles of effective prompting.
A strong AI workflow generally benefits from:
- A clear objective
- Relevant context
- Defined constraints
- Appropriate tools
- Structured outputs
- Evaluation criteria
- Iteration
- Human verification
These are the same principles covered in Promptenza’s Prompt Anatomy guide.
You can also learn the broader methodology in Prompt Engineering: Complete Guide.
Could AI Agents Discover More Biological Systems?
Potentially, yes—but this remains an area of active research.
Anthropic says its life-sciences group wants to apply the approach to additional problems in genomics and other areas of biology.
The broader opportunity comes from the enormous quantity of biological data now available.
Researchers have access to increasingly large genomic, protein and molecular datasets. The challenge is finding meaningful patterns within them.
AI agents could become useful for narrowing those search spaces.
But their usefulness will depend on factors such as:
- Data quality
- Model reliability
- Search methodology
- Tool access
- Scientific validation
- Reproducibility
- Human expertise
In other words, more autonomous AI does not eliminate the need for scientists—it changes where scientists may spend their time.
The Bigger AI Story Isn’t Just the Enzyme
The ART finding is interesting because of the biological system itself.
But from an AI perspective, the more significant story may be the workflow.
A relatively small number of scientists can use AI agents to investigate a search space containing enormous quantities of biological information.
That could change how researchers approach problems where the limiting factor isn’t necessarily a lack of data, but the human capacity to examine it.
The Claude experiment provides an early example of that model.
And it raises a much bigger question:
Could AI agents eventually become a standard part of scientific discovery?
The answer isn’t established yet.
But experiments like this show why agentic AI is moving beyond simple chatbot interactions.
Frequently Asked Questions
What did Claude discover?
Anthropic says Claude agents identified a previously uncharacterized biological system associated with reverse transcriptases and repeating DNA sequences in bacteriophages. The company named the system array-associated reverse transcriptases, or ART.
What is ART?
ART stands for array-associated reverse transcriptases. Anthropic describes the system as consisting of a reverse transcriptase, a partner gene and an array of repeating DNA sequences.
Is ART the same as CRISPR?
No. ART is not currently established as a CRISPR system. Its repeat structure has some similarities to CRISPR-associated arrangements, but its biological function remains under investigation.
Did Claude perform the laboratory experiments?
No. Claude agents performed the computational research and generated candidates and hypotheses. Human scientists performed the laboratory experiments.
Can ART edit genes?
There is currently no established evidence that ART can be used as a gene-editing technology.
How long did the Claude research campaign take?
Anthropic’s technical report describes approximately 21.5 hours of autonomous agent activity involving 949 sessions and more than 215 million tokens.
How many biological sequences did the agents examine?
Anthropic says the agents gathered more than 200,000 reverse transcriptases during the investigation. The technical report describes the wider search across approximately 1.9 billion protein clusters.
Why is the discovery important?
It provides an early example of AI agents helping researchers search massive biological datasets, identify an unusual pattern and generate a candidate system for experimental investigation.
Test Your AI Research Knowledge
1. What does ART stand for?
A. Artificial Research Technology
B. Array-Associated Reverse Transcriptases
C. Automated RNA Technology
D. Advanced Repeat Technology
Answer: B
2. Where was the ART system identified?
A. Human DNA
B. Plant genomes
C. Bacteriophage-associated genetic material
D. Computer code
Answer: C
3. What is a reverse transcriptase?
A. An enzyme that can copy RNA into DNA
B. A type of AI agent
C. A DNA sequencing machine
D. A CRISPR protein
Answer: A
4. Is ART currently proven to be a gene-editing technology?
A. Yes
B. No
Answer: B
Explanation: Anthropic says the system’s function is still being investigated.
5. Who performed the laboratory experiments?
A. Claude agents
B. Human scientists
C. An automated robot laboratory
D. Independent AI researchers
Answer: B
Final Takeaway
Claude’s reported enzyme discovery is an important example of how AI agents are beginning to participate in scientific research at a much larger scale.
The most interesting part isn’t simply that Claude recognized a strange DNA pattern.
It’s the workflow behind it:
Massive biological dataset → AI agents → unusual pattern → hypothesis → human experiments → further scientific investigation.
Anthropic’s ART system is still an early-stage finding. Researchers do not yet know its precise function, and there is currently no basis for calling it a replacement for CRISPR or a proven gene-editing technology.
What has been demonstrated is more measured—and arguably more interesting for the future of AI: AI agents can help researchers explore scientific search spaces that would be extremely difficult to investigate manually at the same scale.
As agentic AI becomes more capable, discoveries like this could become an increasingly important part of the story of how humans use AI for science.
Related Promptenza Guides
For readers who want to understand the AI techniques behind this development:
- Prompt Engineering: Complete Guide to Writing Better AI Prompts
- Prompt Anatomy: Parts of an Effective AI Prompt
- AI Prompting Techniques
- Claude Prompting
- AI Agents and Autonomous AI Workflows
- AI Glossary
Last Updated: 28th September, 2026
