Anthropic says its Claude AI helped uncover previously unknown enzyme system that may enable new gene editing

US AI startup Anthropic revealed its first scientific breakthrough from its molecular biology lab, a previously unknown enzyme system detected by its Claude AI model. This discovery underscores the potential of AI to accelerate basic research, a field where enzyme systems have historically led to major scientific advancements. Anthropic’s Claude model scrutinized a vast database of DNA sequences for 21 hours, uncovering over 200,000 genes for a specific type of enzyme.

From this massive data set, the AI isolated 20 promising genes for further investigation, a task that would have taken a human scientist weeks or even months to accomplish. The AI agents, which worked independently, were given instructions by Anthropic researchers and then conducted their own experiments to determine which leads to pursue.

One of these agents discovered a peculiar arrangement of DNA sequences in bacteria-infecting viruses. This arrangement consisted of a series of short, repeating DNA segments near two genes—one encoding a known enzyme and the other a mystery protein. This configuration mirrors the structure of CRISPR, a bacterial defense mechanism that scientists have repurposed as a gene-editing tool.

Anthropic’s CEO, Dario Amodei, highlighted this finding during a recent address to the UN Security Council, moderated via video link. He described the discovery as preliminary but potentially representing a new gene-editing mechanism with applications in gene therapy. Amodei suggested that this could lead to "profound progress" in treating diseases that have plagued humanity for generations.

However, the precise function of this newly discovered system, dubbed ART, remains unclear. The company also mentioned that their initial experiments suggest the system may be programmable, similar to CRISPR. The findings, published as a preprint, have yet to be reviewed by other scientists. Feng Zhang, a prominent CRISPR pioneer and professor at MIT and the Broad Institute, praised the identification of the repeated DNA fragments as "genuinely intriguing" and called for further investigation.

Anthropic’s lab team, formed last spring, operates in the San Francisco Bay Area, focusing solely on human-operated research and avoiding any experiments involving pathogens that could infect humans. While Amodei did not rule out the possibility of Claude eventually controlling lab equipment under appropriate safeguards, he emphasized that the AI agents conducted the initial discovery autonomously.