Recently, Anthropic revealed the existence of a wet biology lab in the Bay Area, where it utilizes its AI models for hands-on experiments. This week, the organization announced a noteworthy finding: a new enzyme “system” featuring characteristics similar to CRISPR, according to Anthropic.
CRISPR is a naturally occurring immune mechanism in bacteria that helps combat viruses, and it has evolved into a widely used gene-editing technology.
Anthropic claims to have discovered an enzyme system previously unknown, embedded within the DNA of bacteriophages—viruses that infect bacteria and reproduce within them. The researchers indicated that this system can perform tasks akin to CRISPR, such as cutting, copying, and pasting DNA.
The validity and significance of this finding remain to be assessed by the broader scientific community. Anthropic’s CEO Dario Amodei acknowledged that the discovery builds upon earlier work by others, mentioning a Stanford research team that found a system “that is in some ways similar to the one Claude found.”
Nevertheless, Amodei and Anthropic are keen to highlight that this discovery was primarily achieved through their AI model, Claude. Remarkably, the lab, which opened earlier this year, managed this groundbreaking finding in an impressively short time; the team stated that it took Claude just 21 hours of focused analysis with around 950 agents processing 210 million tokens.
This announcement comes at a time when AI leaders, including Amodei, have acknowledged the growing capabilities and risks of AI, suggesting the need for the industry to establish safety measures. Earlier comments by some of his employees raised concerns about the potential existential threats posed by AI.
Amodei has expressed particular concern regarding the potential misuse of AI in bioterrorism. However, he also envisions that AI could potentially “cure most diseases in 5-10 years.” Thus, it seems Anthropic believes that the benefits outweigh the risks.
Interestingly, the most significant aspect of this story may not be the discovery itself, but rather the fact that Claude was not directly involved in the hands-on experiments, which were performed by human scientists.
“Our lab resembles a standard molecular biology facility. We conduct research that falls under the lower biosafety risk levels (BSL-1 and BSL-2) and do not engage with pathogens that can infect humans. All experiments are carried out by human scientists,” they stated.
Moreover, AI-driven biological research is not exclusive to Anthropic. A Stanford team recently published findings on their work combining LLMs with CRISPR, while researchers at UC San Francisco have utilized AI to design enzymes from scratch. Google also introduced its AI-driven biological research platform, AlphaFold, in 2020, indicating that AI has indeed gained a foothold in the field of biological research.
Although Amodei hasn’t dismissed the idea of a fully automated lab in the future, he clarified that currently, “it may even be possible for Claude to autonomously conduct experiments with the right safety measures, but we are not doing that at the moment.”
