Anthropic says its newly established biology lab has identified a previously unknown enzyme system that could have properties similar to CRISPR, in a discovery the company says was made largely with the help of its Claude AI models.
Anthropic confirmed last week that it operates a wet biology laboratory in the Bay Area, where its AI systems are used to support physical biological research.
The company has now announced that researchers using Claude identified an enzyme system hidden in the DNA of bacteriophages, viruses that infect bacteria.
Anthropic says the system has properties reminiscent of CRISPR and may be capable of carrying out operations involving DNA, including cutting, copying and pasting.
What did Claude identify?
CRISPR is a natural defence mechanism used by bacteria against viruses. Scientists have adapted CRISPR into a widely used gene-editing technology.
According to Anthropic, the newly identified system was found in bacteriophage DNA and appears to have some functional similarities to CRISPR.
However, the broader scientific community will need to independently validate the discovery and determine how novel and significant the system is.
Anthropic CEO Dario Amodei acknowledged that the finding builds on previous research, noting that Stanford researchers had previously identified a system with some similarities to the one identified by Claude.
Claude analysed millions of tokens
Anthropic says the discovery was made “mostly, though not entirely, by Claude”.
According to Amodei, Claude spent around 21 hours conducting the analysis, searching through biological data using approximately 950 agents and processing about 210 million tokens.
Anthropic’s biology lab was established earlier this year, making the reported timeline notable given the relatively short period involved.
The company has not disclosed exactly how many months the laboratory has been operating.
Human scientists still conduct experiments
Despite the role of AI in identifying the enzyme system, Anthropic says Claude did not independently carry out the physical laboratory experiments.
The company says its Bay Area facility operates at the lower biosafety levels, BSL-1 and BSL-2, and does not handle pathogens capable of infecting humans.
“All of the lab work is performed by human scientists,” Anthropic said.
The distinction is significant because Amodei has previously discussed both the potential benefits and risks of increasingly capable AI systems in biological research.
He has identified the potential use of AI in bioterrorism as a major concern while also arguing that advanced AI could contribute to major medical breakthroughs.
AI is already being used in biology
Anthropic’s work is part of a broader trend of AI being applied to biological research.
Researchers at Stanford have been exploring the use of large language models alongside CRISPR and gene-therapy research, while scientists at the University of California, San Francisco have used AI to help design enzymes.
Google DeepMind’s AlphaFold has also been used to predict protein structures, demonstrating how AI has become increasingly integrated into biological research.
Anthropic says its approach currently keeps humans responsible for physical laboratory work.
Amodei has nevertheless suggested that fully automated biological research could eventually become possible if adequate safeguards are established.
“Eventually it may even be possible for Claude itself to safely perform the experiments by autonomously controlling lab equipment, with appropriate safeguards in place, but we aren’t doing that today,” he said.
