Anthropic announced that its molecular biology lab in the San Francisco Bay Area identified a previously unknown enzyme system inside the DNA of bacteriophages, viruses that infect bacteria and replicate within them. The company says the system has properties reminiscent of CRISPR mechanisms, as it can perform operations such as cutting, copying, and pasting DNA.
CRISPR occurs naturally as part of an immune system in bacteria against viruses and later became a widely used technology for gene editing. Therefore, finding a biological mechanism that performs similar functions could be important, but the discovery’s degree of novelty and scientific value have not yet been determined.
Claude’s Role in the Discovery
According to Dario Amodei, Anthropic’s CEO, the discovery was made “mostly, though not entirely” by Claude. The model used about 950 agents to process the data, and the process consumed 210 million tokens, while the intensive analytical work took only 21 hours. Anthropic did not specify how many months had passed since the lab began operating, but said it was established in the spring of this year.
At the same time, Amodei acknowledged that the result was built on the work of other researchers, noting that a team from Stanford had previously discovered a system similar in some respects. The broader research community will have to verify the discovery’s scale and originality, rather than treating the company’s description of it as a final result.
What the Model Did Not Do
The material’s most significant revelation is not the discovery alone, but the limits of Claude’s autonomy inside the lab. Anthropic confirmed that all laboratory work was carried out by human scientists, and that the lab operates under biosafety levels BSL-1 and BSL-2 and does not handle pathogens capable of infecting humans. According to the available information, Claude’s role was limited to searching the data and helping reach the enzyme system, not autonomously controlling equipment or conducting physical experiments.
Why Does This News Matter?
The development comes as leaders in the artificial intelligence sector warn of the risks of using models in sensitive biological fields, including the possibility of misuse in bioterrorism. At the same time, Amodei believes that artificial intelligence could help treat most diseases within 5 to 10 years. This reveals a clear trade-off between accelerating scientific research and the need for strict safety controls.
Anthropic is not the only organization using artificial intelligence in biological research; researchers from Stanford have published work on language models and CRISPR, researchers from the University of California, San Francisco have used artificial intelligence to design enzymes from scratch, and Google launched the AlphaFold tool for biology research in 2020.
Anthropic does not rule out eventually creating a lab in which Claude can autonomously control laboratory equipment, but says that this is not happening today and that any transition from analytical assistance to independent experimentation would require appropriate safeguards.