Anthropic Claude Agents Discover CRISPR-Like Enzyme

Anthropic has announced that its newly established wet lab has achieved its first major milestone by autonomously discovering a novel enzyme system. Using a swarm of nearly 950 Claude agents, the AI system analyzed a massive database of DNA sequences to identify a biological mechanism that the company compares to the gene-editing tool CRISPR. The discovery represents an early validation of Anthropic's push to apply its large language models directly to scientific research.

The autonomous discovery process took the Claude agents 21 hours to complete, during which they processed approximately 210 million tokens. Human scientists at Anthropic limited their involvement to writing the initial prompt and performing the subsequent laboratory validation. During the run, one of the AI agents spotted an unusual repeating pattern in the genetic dataset, which was flagged for human review. Further analysis and wet lab testing confirmed the presence of a previously uncharacterized enzyme system found in bacteriophages, which are viruses that infect bacteria.

For biological researchers and AI practitioners, this development demonstrates the viability of using multi-agent AI systems to conduct automated hypothesis generation and data mining at scale. While Anthropic admits it is still working to understand the exact function of the newly discovered enzyme system, the workflow proves that LLMs can identify complex biological structures without constant human oversight. This approach could significantly accelerate the early stages of drug discovery and genetic research by offloading the initial search phase to autonomous AI agents.

The announcement comes at a critical time for Anthropic as the company prepares to go public and seeks to recruit more scientific talent to its laboratory. By demonstrating concrete scientific outputs, Anthropic hopes to position its models as essential tools for the biotechnology sector. This strategy contrasts with competitors like OpenAI, which have recently focused on solving complex mathematical problems, highlighting a growing divergence in how top AI labs seek to prove the real-world utility of their frontier models.