Anthropic announced this week that its newly established biology laboratory has identified a previously unknown enzyme system within bacteriophage DNA. The discovery, which the company describes as possessing properties similar to CRISPR, was primarily driven by its Claude AI model.
Confirmed last week as operating a wet lab in the Bay Area, Anthropic revealed that the team found the enzyme “system” after Claude spent 21 hours analyzing data. The AI utilized approximately 950 agents and processed 210 million tokens to identify the biological mechanism capable of cutting, copying, and pasting DNA.
While CRISPR is a well-known gene-editing tool derived from bacterial immune systems, Anthropic noted that their finding represents a distinct, novel system. However, CEO Dario Amodei acknowledged that the discovery builds upon prior research, including similar work by a Stanford team. He emphasized on social media platform X that the breakthrough was significant largely because of the speed and autonomy with which Claude identified it.
The announcement comes amidst a broader industry conversation about the pace of AI development. Amodei has recently called for slower advancement and improved safety testing, following warnings from some employees about existential risks. Despite these concerns, Anthropic maintains that the potential benefits are substantial. Amodei has previously stated his belief that AI could cure most diseases within five to ten years, suggesting the company views the rewards as outweighing the dangers.
Amodei clarified that current physical experiments in the lab are conducted by human scientists working at Biosafety Level 1 and 2, handling no pathogens infectious to humans. Full automation, where AI might control lab equipment directly, remains a future possibility rather than a current practice.
This development is not isolated to Anthropic. Research institutions such as Stanford and the University of California, San Francisco, are also leveraging AI for biological research, including CRISPR applications and de novo enzyme design. Google DeepMind’s AlphaFold, launched in 2020, pioneered the use of AI in structural biology years ago.
As AI continues to integrate into biological research, the distinction between AI-assisted discovery and fully autonomous scientific exploration will likely become a focal point for both innovators and regulators.
Novel CRISPR-like systems are huge for gene therapy. Can’t wait to see clinical applications.
Does anyone else find it ironic they want slower AI development while building labs this fast?
21 hours versus decades of traditional research. The gap is widening terrifyingly fast.
Fascinating speed, but I wonder how much of this was already in the training data rather than true novel discovery.