Artificial intelligence (AI) is opening up new possibilities in genetic research, but it also raises important bioethical questions regarding the limits of its role in science, the attribution of discoveries, and the need for caution when translating its findings into biomedical applications. The announcement of a potential new CRISPR-like system identified through AI agents brings a fundamental issue back to the forefront: the technological capacity to explore and edit the genome must be accompanied by scientific rigor, responsibility, and a proper ethical assessment of its possible applications. At present, the finding requires experimental validation and serves as a reminder that, while AI can be an extraordinary tool to assist researchers, it replaces neither scientific verification nor human responsibility in the use of biotechnology.
Since the Renaissance, scientific research has relied on experimentation as a means of unraveling the causes of natural phenomena and establishing cause-and-effect relationships. The Oxford English Dictionary defines the “scientific method” as “a method consisting in systematic observation, measurement, and experiment, and the formulation, testing, and modification of hypotheses.” This is how we advance knowledge. When it comes to explaining natural phenomena, nothing compares to experimental demonstration.
This begins with a hypothesis that seeks to explain the phenomenon under study; a methodology and materials are then selected, and experimental tests are conducted to test the hypothesis. The experimental method was redefined in the 20th century by Austrian Karl Popper (1902-1994), who, in addition to demonstrating the veracity of a hypothesis through experimentation, added the requirement to prove that it is not false. According to Popper, only that which can be shown to be both true and unfalsifiable is an object of scientific inquiry. If the facts are as demonstrated by experimentation, then repetition of the same experiment under the same conditions must lead to the same conclusions.
An experimental finding cannot be considered scientifically validated until it has been confirmed that nothing contradicts it. This requirement is what gives science its strength, and explains why this field of human knowledge has become essential to achieving many of the technological and social advances that contribute to the well-being of humanity. The French biologist and geneticist François Jacob (1920-2013), Nobel Prize winner in Medicine in 1965 for his research on the regulation of gene expression, said that science consists of “a constant dialogue between imagination and experiment that allows one to form an increasingly fine-grained conception of what is called reality.”
There is also a final and necessary element in the scientific process: the publication of proven facts. In today’s knowledge-based society, publishing research in a scientific journal—subjected to the scrutiny of independent experts through a peer review process—is considered an indispensable prerequisite for its acceptance by the scientific community.
Distinguishing between exploration and experimentation
Today, we use the term “experimental mathematics” to refer to approaches based on mathematics and statistics, or more specifically, on computer science, computing, and numerical experimentation, to visualize hidden patterns, test conjectures, analyze images, and suggest new paths toward theoretical proof. The result of this progress is artificial intelligence (AI). In reality, there is no intrinsic difference between what a mathematician can do with a pen and paper and what a computer does; the big difference lies in the time it takes to solve a problem.
Next, we will look at an example of how the formal sciences support the experimental sciences in the field of gene therapy using the CRISPR-Cas9 gene editing system.
The fruits of experimentation. The CRISPR-Cas9 system
Many of the discoveries that have revolutionized biology and medicine have originated from the observation of something unusual amid the amazing diversity of molecular phenomena found in nature.
This is the case with “restriction enzymes”, proteins that recognize and cut DNA at specific sites. These types of molecules were discovered in the early 1960s in bacteria and act as a defense mechanism against the DNA of the invading viruses that infect them, namely bacteriophages. Werner Arber, Hamilton Smith, and Daniel Nathans were awarded the Nobel Prize in Medicine in 1978 for their discovery. The isolation of these enzymes allowed researchers to use them to cut different types of DNA, contributing to the development of so-called genetic engineering and, along with it, to the isolation of DNA sequences, transgenesis, DNA sequencing, genomics, etc.
Similarly, the discovery of the enzyme Taq polymerase—a thermostable enzyme isolated in 1976 from the bacterium Thermus aquaticus that copies DNA at high temperatures—was key to its use in the polymerase chain reaction (PCR) technique, a method of controlled DNA replication developed in 1983 by American biochemist Kary Mullis, winner of the 1993 Nobel Prize in Chemistry. This technique revolutionized molecular genetics and is widely used in much of modern diagnostics.
Likewise, the technology known as CRISPR-Cas9 is the result of experimental research in molecular biology and genetics. Its discovery is credited to Francisco Juan Martínez Mojica, a microbiologist from Elche in Spain, who in the 1990s observed a region in the genome of single-celled microorganisms—namely archaea (Archaea)—composed of short, regularly repeated sequences separated by spacers, which he called CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats).[1]
The repeat sequences of the CRISPR complex, in combination with the Cas9 endonuclease enzyme, are found in many bacteria and archaea and constitute a defense system against the invasion of foreign DNA from bacteriophages or other agents. It is a system for degrading invading DNA. After the invasion, the foreign DNA is fragmented, and the resulting pieces are integrated into the bacterial genome among the short repeat sequences of the CRISPR system. When the same invading agent attacks again, these foreign DNA fragments integrated into the CRISPR region are transcribed. In other words, they give rise to RNA molecules that act as guides, recognize the invading DNA and, after recruiting the Cas9 enzyme, degrade it.
The remarkable discovery of an immune system in the microbial world immediately prompted efforts to develop applications for cutting or degrading any type of DNA with a known sequence. Today, CRISPR is an excellent tool for performing “gene therapy”, among other applications. The combination of a synthetic RNA and the Cas9 enzyme is used as “molecular scissors”, a high-precision tool to cut DNA at a specific site in the genome of any organism and then correct it.
The breakthrough came in 2012, when researchers Emmanuelle Charpentier and Jennifer Doudna described how to use CRISPR to “edit” genes involved in human genetic diseases.[2] For this contribution, they received the Princess of Asturias Award for Scientific and Technical Research in 2015 and the Nobel Prize in Chemistry in 2020. Since then, owing to its speed, precision, and cost-effectiveness, the CRISPR-Cas9 system has been adopted by laboratories worldwide to modify DNA for a number of purposes and across multiple species.
AI and the alleged discovery of a new gene editing system
This finding has been pre-published on the alphaXiv platform—although it has not yet undergone peer review[3]—and has been accompanied by extensive media coverage.
According to Anthropic’s announcement, the initial experiment involved the use of around 950 AI agents. Broadly speaking, these agents are autonomous software systems or intelligent digital assistants designed to perform tasks and make decisions with some degree of independence. In just 21 hours, Claude agents scanned a massive database of genome sequences from jumbo phages. In analyzing the DNA adjacent to genes for reverse transcriptase (RT) enzymes (enzymes that synthesize DNA from RNA), the agents identified the same short sequence of DNA letters, repeated over and over again. This is reminiscent of CRISPR arrays.
Claude agents gathered over 200,000 RTs, selected 3,500 new candidate systems, and narrowed the selection down to the 20 most promising ones, which they analyzed to produce readable reports. What Anthropic found in 21 hours would have taken an expert scientist weeks or even months of work.
The regions found were named array-associated reverse transcriptases (ART). They appear as discrete units during Staphylococcus phage infection, suggesting the existence of an RT system directed by a repertoire of different RNAs with functions similar to those of the CRISPR system. The discovery of ART repeats and their association with RNA synthesis sites is reminiscent of the mechanism by which CRISPR generates guide RNAs. While analyzing the DNA sequence near the RT, Claude exclaimed: “[The DNA next to the RT] is spectacular: I can see by eye a tandem repeat array… that’s a CRISPR-like… repeat array?!”
These findings suggest that while AI language models can help guide analyses that open up new avenues for biological research, they do not mean that a new gene editing system has already been discovered.
It should be noted that CRISPR is not simply “a repeat sequence”. It is much more than that: it is a set of mechanisms that allow the recognition and modification of a specific target, such as the DNA of a virus or an invading agent. None of this follows from Anthropic’s findings.
It should also be clarified that although the similarity to CRISPR may be useful as a reference, it may also be misleading: the main similarity is in the genetic architecture. Anthropic expressly acknowledges that its exact biological function is not yet known and has not been shown to enable DNA editing.
The news coverage is marred by a degree of sensationalism
The news has been broadcast through The New York Times and has been reported worldwide by various media outlets in a sensationalist tone. In some cases, it has been taken for granted that we are looking at a new system comparable to CRISPR, even though this is not yet supported by empirical science. At present, it is simply a finding that needs to be confirmed experimentally. In order for it to be considered a new gene editing tool similar to CRISPR, researchers would have to demonstrate, among other things, which molecule it recognizes; which sequence it targets; which protein carries out the modification; whether it cuts, copies, inserts, or modifies DNA; whether it can be programmed; how precise it is; and whether it works outside the organisms in whose genome it was discovered. None of this has yet been established.
However, an even more interesting question arises: to whom should the discovery be attributed?
The article mentions Spanish researcher Mario Rodríguez Mestre, a scientist at the University of Copenhagen, who says that for years he and his colleagues had been studying the microbial system that Anthropic claim to have discovered, and that some of the genes and bacteriophages overlapped with their research. Mestre also said that he had uploaded information about the system to Anthropic’s AI models. He himself acknowledges that he has no evidence that Anthropic used his unpublished work, so this does not demonstrate any appropriation. Anthropic, meanwhile, maintains that it was not aware of any published research describing the same finding, and that Claude was not trained on user chats.
These similarities highlight the delicate role of AI systems whose scope depends on algorithms capable of processing vast amounts of data, while also raising questions about the provenance and traceability of that data. Of course, these issues do not arise in quite the same way in laboratory-based experimental research.
Several researchers have warned that portrayals of Anthropic’s finding as a major AI discovery are exaggerated. Some press headlines have claimed that AI discovered the system “on its own”, while others have declared that “AI discovers an alternative to CRISPR”, or branded it a “100% autonomous discovery”, or an “Immediate revolution in medicine”.
All this is overstated and, moreover, clouds how the finding actually occurred. AI cannot think for itself or conceive a new biological concept. The main credit belongs to the human researchers who defined the research framework, created the prompts, equipped the computers with algorithms and training, and obtained digital empirical results.
If AI examines a human-built database, uses human-created scientific literature, employs human-developed concepts, and then proposes a hypothesis that other humans have to verify experimentally, it may be more accurate to describe it as an AI-assisted discovery rather than a discovery by AI, with the sensationalism that usually accompanies this fourth industrial revolution technology.
It is evident that Anthropic has the ability to explore millions of genomic data points in search of molecular tools in just a few hours. The really interesting news is that AI has found—within a massive amount of genomic information—a biological pattern that researchers consider sufficiently unusual and interesting to warrant further investigation in the laboratory. This is precisely one of the roles AI can play: opening up new avenues of research that must now undergo experimental validation. If we end up demonstrating that ART can be programmed to manipulate DNA, then we will indeed be looking at something extraordinary. However, we are still a long way from that.
In matters of such importance, we must move forward step by step and refrain from celebrating prematurely. AI can greatly assist as a tool for scientific exploration, but it is not sufficient on its own.
There is no evidence that the repeats found by Anthropic’s team of AI agents perform functions similar to those of the CRISPR system. The most important part remains to be proven: that this system can compete with CRISPR and become a new biotechnological tool. At the moment, this possibility is purely speculative; what matters most is putting each supporting procedure in scientific research in its proper place. We must not overstate the role of AI, as has happened in this case and happens daily in many others, especially in the media.
Ultimately, AI is still a working tool, with a remarkable ability to analyze millions of data points, find similarities, and even suggest functions for its findings. Nevertheless, once AI makes a suggestion, the crucial step remains: testing it in the laboratory.
That being said, from an ethical perspective, this finding will hopefully be confirmed so that, in the not-too-distant future, a second technology for editing and repairing DNA will become available that is at least as effective and promising for its application in gene therapy as CRISPR has proven to be.
Nicolás Jouve
Professor Emeritus of Genetics at the University of Alcalá
Member of the Bioethics Observatory of the Catholic University of Valencia
References:
[1] Mojica FJ, Ferrer C, Juez G, Rodríguez-Valera F. “Long stretches of short tandem repeats are present in the largest replicons of the Archaea Haloferax mediterranei and Haloferax volcanii and could be involved in replicon partitioning”. Mol. Microbiol. 17 (1995), 85–93.
[2] Doudna JA., Charpentier E. “Genome editing. The new frontier of genome engineering with CRISPR-Cas9”.Science, 346 (6213) (2014).
[3] Yoon PH. Athukoralage J S. Ameisen, E. et al. “Autonomous AI Agents Discover Reverse Transcriptases with Tandem Repeat Arrays”. alphaXiv 2026. https://www.alphaxiv.org/abs/2609.ai-agents-discover-reverse-transcriptases.
