Anthropic revealed it has established a wet biology lab in the Bay Area that harnesses its AI models to assist in physical experiments. Remarkably, this lab has already made a significant finding: a previously unknown enzyme system within bacteriophage DNA that exhibits functions similar to CRISPR, the well-known gene-editing tool used widely in research for cutting and modifying DNA. This discovery, accomplished in a short time frame, is being positioned as a potentially important breakthrough, though broader scientific validation is still required.

The new enzyme system was found through the efforts of Anthropic's AI, Claude, which analyzed massive datasets with about 950 agents processing 210 million tokens over 21 hours. While Anthropic emphasizes Claude’s key role in identifying this biological system, the CEO, Dario Amodei, acknowledged that related systems had been previously discovered by researchers at Stanford, indicating the discovery builds upon existing scientific foundations.

This development comes at a time when AI leaders, including Amodei, have called for a slowdown in AI progress to address safety concerns, especially given fears about AI’s potential misuse in fields like bioterrorism. Despite these risks, Anthropic is confident about the benefits AI could bring to healthcare, including curing many diseases in the next 5 to 10 years. Notably, while AI guided the discovery process, all physical experiments have been conducted by human scientists, with no fully autonomous lab work currently in place.

Anthropic’s venture into AI-driven biological discovery is part of a broader trend, with other institutions like Stanford and UCSF also leveraging AI in protein and enzyme research, while companies such as Google have long been involved through projects like AlphaFold. Looking ahead, Amodei has not ruled out the possibility of an AI autonomously performing lab experiments in the future, though safeguards and incremental steps remain priorities for now.