Artificial intelligence (AI) company Anthropic announced on September 23 that it has discovered a harmful enzyme system with characteristics similar to CRISPR using its AI model, Claude. While Claude explored data and proposed research hypotheses, actual biological experiments were conducted by human researchers.
According to reports from U.S. tech media outlets like TechCrunch on September 24, Anthropic recently revealed that it operates a biological research lab in the San Francisco Bay Area, which opened last spring and has produced its first research results within just a few months.
The research team at Anthropic identified an unknown enzyme system present in bacteriophage DNA. Bacteriophages are viruses that invade and multiply within bacteria. The researchers explained that this enzyme system can perform functions similar to CRISPR, such as cutting, copying, or inserting DNA.
CRISPR is a natural immune system that bacteria use to respond to viruses and is currently utilized as a precise gene-editing technology for correcting specific DNA sequences.
According to Anthropic, Claude mobilized approximately 950 AI agents to analyze the data, utilizing a total of 210 million tokens. The intensive research was conducted over about 21 hours.
Dario Amodei, CEO of Anthropic, stated, "Most of this discovery was achieved through Claude." He noted that the significance and novelty of this finding will need to be validated by the academic community. Amodei added, "This research builds on existing studies, and we only found some similarities with the enzyme system previously discovered by Stanford researchers."
TechCrunch and other media outlets predict that this announcement will spark discussions about the potential and safety of AI in biological research. Amodei has previously identified the risk of AI being misused for bioterrorism as a major concern. At the same time, he has expressed optimism that AI could contribute to treating most diseases within the next 5 to 10 years.
Anthropic clarified that it did not allow AI to directly operate experimental equipment during this research, and all physical experiments were conducted by human researchers. The Anthropic lab only conducts research at biosafety levels BSL-1 and BSL-2. BSL-1 is the lowest biosafety level, dealing with unknown materials that do not consistently cause disease in healthy adults. BSL-2 refers to biosafety levels that handle materials with moderate potential hazards to humans and the environment.
Meanwhile, the use of AI in biological research is already spreading across academia and industry. Researchers at Stanford University have previously published results combining large language models (LLMs) with CRISPR. Researchers at the University of California, San Francisco (UCSF) have also designed enzymes from scratch using AI. Google unveiled its AI model, AlphaFold, for predicting protein structures in 2020.
Anthropic has also opened the possibility of a fully automated research lab where AI could autonomously control experimental equipment in the long term. Amodei remarked, "With appropriate safety measures, it may be possible for Claude to autonomously control experimental equipment and conduct experiments in the future, but we are not doing that at this time."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.