Anthropic claims AI model discovers new enzyme system in biology lab

Anthropic has announced that its artificial intelligence model, Claude, has identified a previously unknown enzyme system with properties similar to CRISPR. The discovery was made within the company’s wet biology laboratory in the Bay Area, where AI models are used to guide physical experiments. The AI giant describes the finding as a significant breakthrough, noting that the new system can perform operations such as cutting, copying, and pasting DNA, much like the natural immune system used by bacteria to fight off viruses.

The enzyme system was hidden in the DNA of bacteriophages, which are viruses that infect and replicate within bacteria. While the broader research community will need to validate the novelty and significance of this discovery, Anthropic emphasises that the work was conducted primarily by its AI. CEO Dario Amodei stated that the discovery was found mostly, though not entirely, by Claude. He acknowledged that the work built upon previous research, including a similar system discovered by a team from Stanford, but highlighted the speed and efficiency of the AI-driven process.

The laboratory was established earlier this spring, and Anthropic declined to specify exactly how many months it has been operational. Despite the short timeframe, the company reports that Claude required only 21 hours of concerted effort to make the discovery. During this period, the AI searched through data using approximately 950 agents and processed 210 million tokens. This rapid pace of discovery stands in contrast to traditional biological research timelines, suggesting a potential shift in how scientific breakthroughs are achieved through the integration of advanced AI tools.

The announcement comes at a sensitive time for the artificial intelligence industry. Recently, several AI executives, including Amodei, have publicly admitted that current models have become so capable and potentially dangerous that the sector must slow down to develop robust safety-testing procedures. Earlier this month, some of Anthropic’s employees publicly expressed concerns that there is a risk AI could lead to catastrophic outcomes. Amodei himself has stated that one of his greatest fears is the use of AI for bioterrorism. However, he also believes that AI will cure most diseases within five to ten years, indicating that Anthropic views the potential rewards as outweighing the risks.

Despite the AI-driven discovery, the physical experiments were not conducted by the model. Anthropic clarified that its Bay Area lab operates at lower biosafety risk levels, specifically BSL-1 and BSL-2, and does not handle pathogens that can infect humans. All laboratory work is performed by human scientists. This distinction is crucial, as it highlights that while AI can identify and design biological solutions, the physical execution remains a human task. The company has not ruled out the possibility of a fully automated biology lab in the future, with Amodei suggesting that Claude may eventually control lab equipment autonomously with appropriate safeguards, though this is not currently the case.

The integration of AI into biological research is not unique to Anthropic. Researchers at Stanford have recently published work on using large language models with CRISPR, while scientists at UC San Francisco have used AI to design enzymes from scratch. Google launched AlphaFold, its first AI-powered biology research tool, in 2020. These developments indicate that AI has firmly established itself in biological research beyond the major AI laboratories. As the industry continues to advance, the balance between rapid innovation and safety remains a central concern for both the public and the developers of these powerful technologies.