Nearly 950 Claude AI agents searched a vast DNA database for 21 hours before identifying a previously uncharacterized biological system with features reminiscent of CRISPR, though Anthropic says scientists still do not know what the system does or whether it will have practical gene-editing applications.
By yourNEWS Media Newsroom.
Anthropic says its Claude artificial intelligence has made the first major discovery to emerge from the company’s new biology laboratory, identifying an unusual enzyme system that shares intriguing characteristics with molecular systems used for programmable gene editing.
The finding is preliminary, and researchers do not yet know the biological function of the newly identified system. But Anthropic is presenting the experiment as an early demonstration of how AI agents could independently comb through enormous biological datasets, recognize previously overlooked patterns and generate discoveries for human scientists to investigate.
In an announcement describing the research, Anthropic said scientists gave Claude a high-level assignment to search a massive DNA sequence database for unusual examples of reverse transcriptases, enzymes that convert RNA into DNA.
From there, according to the company, the AI largely took over the computational search.
Roughly 950 Claude agents spent 21 hours processing about 210 million tokens while examining the genomic data. The agents gathered more than 200,000 reverse transcriptases, identified thousands of potential new systems and narrowed the candidates before one agent noticed a highly unusual repeating DNA pattern adjacent to an atypical reverse transcriptase.
The pattern was flagged for human scientists, who conducted additional analysis and physical laboratory experiments.
Anthropic said the work ultimately revealed what it describes as “a previously uncharacterized enzyme system found in bacteriophages,” the viruses that infect bacteria. The company calls the system array-associated reverse transcriptases, or ART.
Anthropic says its scientists’ involvement in the initial discovery process was largely limited to formulating the original prompt and conducting the physical laboratory work. Humans subsequently reviewed Claude’s findings and experimentally investigated the candidate system.
The distinction is significant because Claude did not physically conduct experiments. Anthropic’s scientists performed the wet-lab work necessary to investigate whether the computational finding corresponded to an actual biological system.
What ART actually does remains unknown.
Anthropic says the system includes a reverse transcriptase, another nearby gene whose function is not yet understood and an array of regularly spaced DNA repeats. Those repeats bear structural similarities to CRISPR arrays, which form part of the biological machinery that ultimately gave scientists programmable gene-editing tools.
The resemblance does not mean Anthropic has discovered another CRISPR or demonstrated a new method for editing human genes.
The company said the characteristics identified in ART have previously appeared together in only a small number of other biological systems, all of which are programmable and can perform functions such as cutting, copying or inserting DNA. Laboratory work is continuing to determine whether ART has comparable abilities.
That uncertainty places an important limit on the announcement. Anthropic has identified an unusual biological system that may prove scientifically useful, but researchers have not yet established its primary function or demonstrated that it can be developed into a gene-editing technology.
The timing also gives the announcement significance beyond molecular biology.
Anthropic is preparing for a potentially enormous initial public offering as artificial intelligence companies increasingly attempt to demonstrate that advanced models can perform more than conversational and office tasks. The company’s anticipated listing has been discussed at valuations reaching into the trillions of dollars, although final terms remain unsettled.
Scientific discovery has consequently become another arena in which AI developers are attempting to demonstrate the capabilities of their most advanced systems.
Anthropic formed its life sciences research group in the spring of 2026 and built a Bay Area wet lab where human scientists can experimentally test biological hypotheses generated by Claude. The company says the laboratory handles lower-level biosafety work and does not work with pathogens capable of infecting humans.
Anthropic acknowledged that much remains unknown about its first discovery.
Asked why it was announcing a biological finding before determining what the enzyme system actually does, the company said it is “important to share such findings early, both to demonstrate Claude’s capabilities and to give the broader community insight into what we’re working on.”
The Verge, which examined Anthropic’s announcement, described the disclosure as premature and noted that there is currently no evidence that ART will produce a practical biotechnology application, much less an advance comparable to CRISPR.
The distinction matters because CRISPR transformed genetics only after years of scientific work converted observations about unusual repeating DNA sequences in microbes into a powerful system capable of targeting and altering specific genes.
Gene editing’s potential has also generated significant ethical controversy, particularly when researchers have attempted to alter human embryos.
One of the most notorious cases involved Chinese scientist He Jiankui, who announced in 2018 that his team had used CRISPR to alter human embryos that resulted in the birth of twin girls. The experiment drew international condemnation, and He was subsequently convicted in China of illegal medical practice and sentenced to prison.
Breitbart News reported on the case after He resurfaced publicly following his imprisonment:
According to He, the baby girls were genetically modified to make them highly resistant to HIV, which their father was infected with. Scientists around the world condemned his team for irresponsibly using CRISPR to edit human genes. He attempted to publish papers describing his work, but international scientific journals refused to accept them.
A third allegedly genetically edited child was born to another of the women in He’s trials. None of the three children has been identified to the public, so He’s claims have never been independently verifiable.
The Chinese government was clearly displeased as well, because He disappeared in December 2018, shortly after discussing his work at a conference in Hong Kong.
The case illustrated both the extraordinary capabilities of programmable gene editing and the ethical and safety questions surrounding its application to humans.
Anthropic has not claimed that Claude’s ART discovery can edit human genes, and no such capability has been demonstrated. For now, the comparison with CRISPR concerns similarities in the underlying biological architecture rather than equivalent functionality.
The experiment instead raises a different question: whether AI systems can become useful scientific investigators capable of finding meaningful biological patterns that human researchers have overlooked.
Anthropic says the reverse transcriptase itself had appeared in earlier research. What Claude appears to have recognized for the first time was the broader system surrounding it — including the repeating noncoding DNA sequences and an accompanying protein of unknown function.
That type of pattern recognition is particularly suited to genomic research because scientists now have access to enormous collections of DNA sequences containing countless proteins and biological systems whose functions remain poorly understood.
Traditionally, researchers must search those databases, compare related sequences, examine neighboring genes and determine which anomalies merit expensive laboratory investigation. Anthropic’s experiment suggests that large groups of AI agents may be capable of performing substantial portions of that initial search simultaneously.
Whether those systems can reliably produce important discoveries remains an open question.
The ART finding still requires additional experimentation, and Anthropic has not established that it will become a useful biotechnology platform. The company has released the work early while continuing to investigate how the system functions.
It nevertheless represents a notable test of a rapidly developing model for scientific research: humans defining a broad problem, AI agents autonomously searching enormous bodies of information and human scientists returning to the laboratory to determine whether the machine-generated hypothesis survives physical experimentation.
Anthropic hopes to extend that approach beyond genomics and is already pursuing AI applications in areas including protein engineering and drug discovery. Other artificial intelligence companies are also increasingly using scientific research to demonstrate advanced-model capabilities.
The Verge noted a similar push by OpenAI into advanced mathematics while criticizing aspects of the company’s approach to established academic norms — an assessment reflecting the outlet’s characterization of that work rather than a scientific finding.
For Anthropic, the unanswered question surrounding its newly identified enzyme system may ultimately be as important as the discovery itself.
Claude found something researchers had apparently not characterized before. Scientists must now determine what it does — and whether an unusual pattern found by hundreds of AI agents becomes merely a biological curiosity or the foundation for a useful new molecular tool.
Original article
Comments