Anthropic’s Secretive AI-Powered Wet Lab Breaks Cover and Makes First Discovery

AI giant Anthropic has made several forays into the life sciences over the last year, including through its Claude Science bioinformatics platform. Now, the company has announced that it is moving into wet lab science with the launch of a human-staffed lab in the San Francisco Bay Area. The company said in a release that it had been quietly operating the lab since the spring of this year and that the project has already made a discovery: “a novel enzyme system with properties reminiscent of CRISPR.”

A “Typical” Molecular Biology Lab Powered by AI

In recent weeks, warnings from leaders both inside and out of the tech industry have called for better regulation around AI. Some of these warnings have been explicitly apocalyptic.

Enter Anthropic’s latest life science venture, which is a demonstration of AI being used in “a way that benefits the world,” Anthropic’s head of life sciences Eric Kauderer-Abrams told Reuters. The lab's exact focus and how it works still aren’t completely clear. Kauderer-Abrams said that the venture could target so-called “undruggable” conditions considered too difficult to treat, but a spokesperson for the company later clarified that the lab wouldn’t focus only on drug science.

While Reuters’ unnamed sources said the company’s ultimate aim was an automated setup in which its AI Claude directs robotic lab units, Kauderer-Abrams also told the outlet, "We believe that to do biology, the final test is still and will be for a while in real lab work,” suggesting Anthropic is using a human-first approach, at least for now.

The new venture, said Anthropic in its news release, “looks like a typical molecular biology lab.” The setup’s main difference is upstream of its wet lab. Many of the team’s workflows, Anthropic said, begin by prompting Claude to work its way through DNA sequences linked to poorly understood proteins. It then winnows the list to identify candidate proteins with interesting functions, which a team of human scientists then tests.

AI Discovers ART

Anthropic said that the approach’s first discovery was of a CRISPR-like enzyme system based on reverse transcriptase (RT) enzymes, which write RNA into DNA. A simple prompt to investigate a database of DNA sequences for new RT functions led to a day-long search by 950 AI agents, which amassed over 200,000 candidate RTs. After boiling these down to the 20 most promising contenders, an agent reported that one RT, which had been characterized in prior research, also contained a previously unnoticed tandem repeat array, a repeating set of short DNA code that is integral to the cut-and-paste activity of CRISPR-like systems.¹

In a companion technical report, which has not been peer-reviewed, Anthropic’s scientists called the new system array-associated RTs (ART).² While it isn’t yet clear what ART’s function is, Anthropic’s press release quoted Feng Zhang, CRISPR pioneer and researcher at the Massachusetts Institute of Technology and the Broad Institute. “This is an exciting example of how AI agents can contribute to biological discovery. The identification of RNA-repeat arrays associated with reverse transcriptases is genuinely intriguing and merits further investigating,” said Zhang.

Kauderer-Abrams told Reuters that Anthropic’s research program was treading carefully, mindful of competition concerns. Anthropic said its project would not run clinical trials. “We’re not competing with pharma and biotech companies that make their business in bringing drugs to market,” Kauderer-Abrams said.

But even if Anthropic’s big pharma clients, like Roche, Genentech, and Bristol Myers Squibb, may be reassured by these statements, it’s clear that while the company makes certain AI-powered tools available to the life science community, its own project would have access to big guns beyond the reach of external academics. The Scientist reported in July that some academics were concerned about whether tools like Claude Science would force researchers to choose between funding AI tokens or trainees. ART’s discovery, made with a team of six named human scientists, also used 210 million tokens. “There’s some things that we can do much faster in our own hands,” Kauderer-Abrams said to Reuters.