Recently, Anthropic disclosed that it has established a wet biology laboratory in the Bay Area where it utilizes its AI models for conducting physical experiments. This week, the AI powerhouse revealed that this lab has purportedly made a significant finding: a novel enzyme “system” possessing certain “characteristics akin to CRISPR,” as per Anthropic’s description.

CRISPR represents a natural defense mechanism employed by bacteria to combat viruses, which has transformed into a widely-used gene-editing technology among researchers.

Essentially, Anthropic claims the discovery entails an unidentified enzyme system concealed within the DNA of bacteriophages, viruses that infect and proliferate within bacteria. Researchers at Anthropic assert this system operates similarly to CRISPR in that it can “carry out processes like cutting, copying, and pasting DNA.”

The broader research community will need to assess how substantial or groundbreaking this finding truly is. Anthropic’s CEO Dario Amodei acknowledged that the discovery built upon the efforts of prior researchers, referencing that a Stanford team had previously identified a system “that shares some similarities with the one Claude uncovered,” he tweeted on X.

However, Amodei and Anthropic are not solely focusing on the finding. They are highlighting that it was discovered “primarily, though not exclusively, by Claude,” the CEO noted.

On one hand, this is remarkable. The lab was only launched this spring (although Anthropic chose not to specify how many months it has been operational). Achieving a discovery within such a short timeframe would be impressive; the team mentioned that it took Claude merely 21 hours of dedicated work. Claude sifted through data using approximately 950 agents that processed 210 million tokens.

Nevertheless, the announcement that Anthropic has been engaged in biology experiments follows AI CEOs, including Amodei, publicly expressing that models have grown so powerful, and potentially hazardous, that the industry must decelerate and implement safety testing protocols. This comes after a few of his employees publicly stated earlier this month that AI poses the risk of causing mass harm.

Amodei himself has stated that one of his greatest concerns is that AI could be exploited for bioterrorism. However, he also holds the belief that AI will “eliminate the majority of diseases within 5-10 years,” he wrote. Hence, Anthropic has evidently concluded that the potential benefits outweigh the dangers.

In light of this, perhaps the most significant revelation in this update isn’t the discovery itself, but that the biology lab has not yet unleashed Claude fully. The hands-on experiments were conducted by human researchers.

“Our laboratory, situated in the Bay Area, resembles a conventional molecular biology lab. Our research strictly involves the lower biosafety risk levels (BSL-1 and BSL-2), and we do not engage with pathogens that can infect humans. All laboratory work is conducted by human scientists.”

Moreover, it remains true that AI-enhanced biological research is not solely the domain of Anthropic. Researchers from Stanford recently released a study regarding their efforts with LLMs and CRISPR. Researchers at UC San Francisco have utilized AI to design enzymes from the ground up. Google introduced its first AI-driven biology research instrument, AlphaFold, back in 2020. Thus, AI has evidently penetrated biological research beyond the AI labs actively pursuing it.

Nonetheless, Amodei has not dismissed the possibility of a fully automated biology lab in the future. “Eventually, it might even become feasible for Claude itself to safely conduct experiments by autonomously manipulating lab equipment, with adequate safeguards in place, but we are not implementing that today,” he mentioned.

.