Anthropic says its AI models have helped identify a previously unknown biological system with similarities to CRISPR, marking an early result from the company’s newly established wet lab in the San Francisco Bay Area.
The company says the discovery involves an enzyme system found in bacteriophages — viruses that infect bacteria. According to Anthropic, the system appears capable of manipulating DNA in ways that resemble some of the functions associated with CRISPR, including cutting, copying and moving genetic material.
The finding has not yet been independently validated by the wider scientific community, so its ultimate significance remains to be determined.
Still, Anthropic is highlighting the research as an early example of AI models moving beyond computer-based work and contributing to experiments in the physical world.
Claude Searched Through Millions of Tokens
Anthropic CEO Dario Amodei said the discovery was made largely by Claude, the company’s AI model family, although he acknowledged that the work builds on research conducted by other scientists.
Amodei also pointed to previous work from researchers at Stanford, who identified a biological system with some similarities to the one found by Anthropic.
According to the company, Claude spent about 21 hours working on the research and searched through biological data using roughly 950 AI agents. Those agents collectively processed around 210 million tokens during the investigation.
For a laboratory that was only established earlier this year, Anthropic says the speed of the work is notable.
The company has not disclosed exactly how long the lab has been operating, but its existence was confirmed only recently.
The Experiments Were Still Done by Humans
Despite the emphasis on AI-driven discovery, Anthropic says Claude is not independently conducting physical biology experiments.
Human scientists carried out the laboratory work.
Anthropic says the facility operates at Biosafety Level 1 and Level 2 and does not work with pathogens capable of infecting humans. The company describes the facility as a conventional molecular biology laboratory where human researchers perform the experimental procedures.
That distinction is important because Anthropic has simultaneously been warning about the potential dangers of increasingly capable AI systems.
The company has argued that advanced AI could eventually create serious risks, including the possibility that powerful models could be misused for biological applications.
Amodei has previously identified AI-assisted bioterrorism as one of his major concerns. At the same time, he has argued that AI could dramatically accelerate medical research and potentially help address many diseases within the next decade.
Anthropic’s biology work sits directly at that intersection: the technology could produce significant scientific benefits while also raising questions about how much autonomy increasingly capable AI systems should have in biological research.
What Is CRISPR?
CRISPR is a natural defense mechanism found in bacteria that helps them protect themselves against viral infections.
Scientists have adapted the system into a powerful gene-editing technology that can be used to make targeted changes to DNA. Its applications range from basic biological research to experimental medical treatments.
Anthropic is not claiming that Claude discovered CRISPR itself.
Instead, the company says its researchers identified an enzyme system in bacteriophages that shares certain functional characteristics with CRISPR.
The exact biological significance of the discovery will depend on further research and independent verification.
AI Is Already Changing Biological Research
Anthropic’s experiment is part of a much broader shift toward using artificial intelligence in biology.
Researchers at Stanford have been investigating how large language models can assist with CRISPR-related research, while scientists at the University of California, San Francisco have used AI to design enzymes.
Google has also been developing AI systems for biological research for years. Its AlphaFold project, introduced in 2020, demonstrated how machine learning could help researchers predict the three-dimensional structures of proteins.
Those efforts suggest that AI is becoming increasingly integrated into biological research rather than remaining limited to text generation, programming and other digital tasks.
Anthropic’s work adds another dimension by attempting to use AI agents to search scientific information and develop hypotheses that can then be tested by human researchers.
Could AI Eventually Run the Lab?
For now, Anthropic is keeping humans firmly in control of the physical experiments.
But Amodei has not ruled out a future in which AI systems could operate laboratory equipment themselves.
He said that, with appropriate safeguards, it could eventually become possible for Claude to conduct experiments autonomously by controlling lab equipment.
That possibility would represent a much bigger shift than using AI to analyze scientific literature or suggest experiments. A system capable of independently designing, executing and interpreting biological experiments would give AI a direct role in the physical research process.
Anthropic says that is not what is happening today.
For now, the company’s model is more restrained: Claude searches and analyzes information, proposes or assists with scientific work, and human researchers carry out the experiments.
The reported enzyme discovery is therefore less a demonstration of an autonomous AI scientist than an early test of what can happen when highly capable AI models are placed alongside experienced researchers in a real laboratory.
Whether the newly identified system ultimately proves to be a major biological discovery will be decided by further research. But Anthropic’s experiment already points to a larger change underway in science: AI is increasingly moving from the role of research assistant toward becoming an active participant in the discovery process.
