Artificial intelligence agents associated with the titan Anthropic have scanned genomic data to discover a peculiar pattern of DNA within giant viruses, a finding that the company compared to bacterial CRISPR immune systems. The discovery, released September 23 on the alphaXiv platform and not yet peer reviewed, marks the initial output from Anthropic’s newly launched life sciences research group where human scientists and AI agents collaborate on physical experiments.
Autonomous AI Agents Explore Billions of Proteins
Anthropic’s newly established wet lab enlisted roughly 950 autonomous AI agents running on large-language models to comb through a massive database of DNA sequences encoding billions of proteins. Over a 21-hour period, according to company disclosures reported by The Verge, the agents worked through 210 million tokens in a self-directed fashion. They discussed preliminary results and decided on subsequent steps without human intervention until a pattern emerged. Eric Kauderer-Abrams, head of life sciences at Anthropic, stated that many impactful biological discoveries stem from finding unusual things in microbes, noting that the goal was to systematize and scale up that style of research.
Comparing Viral Repeats to Microbial CRISPR Systems
While examining DNA surrounding reverse transcriptase genes in bacteriophages—viruses that infect bacteria—the AI agents flagged a repeating pattern of short DNA sequences found across multiple giant virus genomes. Bacterial CRISPR systems rely on similar arrays of repeated DNA separated by spacer segments copied from invaders; when an intruder returns, RNA guides a partner enzyme to slice the interloper’s genome. However, researchers have not yet determined what the newly found viral sequences actually do. Kauderer-Abrams called it a promising lead, noting that the real challenge lies in completely characterizing the function and developing it into a useful tool, as there is currently no known DNA-slicing enzyme partnered with these specific viral repeats.
Did You Know?
Anthropic’s first biological experiment involved roughly 950 AI agents spending more than 21 hours exploring billions of proteins autonomously before flagging the uncharacterized enzyme system.
Human Expertise and the Future of Wet Lab Research
The discovery comes as major tech companies increasingly turn to scientific research to prove the practical capabilities of their advanced models. While OpenAI has focused on tackling complex mathematical problems, Anthropic built a physical wet lab because biological research requires running experiments in the physical world, according to Kauderer-Abrams, who emphasized that “biology is hard.” Human involvement in this initial experiment was restricted to providing the initial prompt and performing the necessary physical lab work to verify the findings. Anthropic stated it shared the preprint early to demonstrate the capabilities of its Claude models and give the broader scientific community insight into its drug discovery and life sciences initiatives.
Frequently Asked Questions
What did Anthropic’s AI agents discover?
The AI agents discovered an uncharacterized repeating pattern of DNA in the genomes of giant viruses and bacteriophages that bears structural similarities to the repeat arrays found in microbial CRISPR systems.
Is this new discovery a proven gene-editing tool?
No. Researchers have not yet determined the function of the newfound viral sequences, and there is no known DNA-slicing enzyme partnered with them.
How many AI agents were used in the experiment?
Anthropic deployed roughly 950 AI agents that worked autonomously for over 21 hours to analyze billions of proteins.
Want to stay updated on the intersection of artificial intelligence and life sciences research? Subscribe to our newsletter or explore our latest coverage on emerging biotechnology trends.
Keep reading