Just one week after Anthropic released a report claiming to have shut down potential bioweapons development conducted by unnamed users of its Claude AI, details have now emerged that the AI firm has also been quietly running a robotically automated and AI-operated biology lab itself.
Reuters reporters uncovered the project after speaking with two anonymous sources who explained that Anthropic wants to ramp up Claude AI’s ability to coordinate robotic lab assistants in the performance of science experiments with minimal human intervention. According to Reuters, Anthropic’s new autonomous “wet lab” is operating right now somewhere in the Bay Area. (Seems fine.)
The plan appears to be pitched toward possible drug development. However, it will reportedly stop just short of the clinical trials that might put the firm in direct competition with pharmaceutical industry giants (presumably prospective Anthropic customers), according to Anthropic’s head of life sciences, Eric Kauderer-Abrams, who confirmed the lab’s existence to Reuters.
“We believe that to do biology, the final test is still and will be for a while in real lab work,” Kauderer-Abrams told the wire service. “We absolutely are doing that today.”
In vitro veritas
Kauderer-Abrams went on to emphasize that nothing Anthropic is currently undertaking inside this previously undisclosed lab diverges from “what you would see in most biotech companies.” And there was already some prior indication that the firm was pursuing biotech opportunities.
Earlier this summer, Anthropic touted its protein-design research, both “in silico” (meaning virtual or computation-based only) and “in vitro” (meaning outside complex living organisms, whether inside test tubes, petri dishes, or something else).
But, for years, a desire to spur medical breakthroughs has been a persistent theme from the firm’s CEO, as Amodei has spoken candidly about his father’s tragic death from Hepatitis C just four years before a life-saving medication was first successfully tested on patients. As Amodei put it in some ambitious prognostications back in 2024: “AI-enabled biology and medicine will allow us to compress the progress that human biologists would have achieved over the next 50-100 years into 5-10 years.”
Caveat promptor
Despite those public statements, a spokesperson for Anthropic followed up with Reuters to underline that its new wet lab is not specifically pursuing drug discovery, but declined to get into any details. The firm’s life sciences head framed this more in terms of the project still being in its nascent phases.
“We’re in the very early innings of using AI to automate the execution of lab work,” Kauderer-Abrams said, calling it “an area that has the potential to bring about meaningful acceleration in so many different processes.”
While the debate on the safety and prudence of AI-assisted biotech R&D will likely stretch on years or even decades, it’s been a remarkably confusing week for the discourse. Last Saturday, Anthropic CEO Dario Amodei announced that companies like his own should ‘pause’ further AI development in the face of myriad credible-to-alarmist concerns about the technology’s power to end human civilization as we know it.
Soon thereafter, like addicts in the bargaining phase of denial, Amodei and his AI competitors appeared to land on the concept of “pacing the frontier” as a safety-conscious slowdown in AI research that doesn’t disrupt the precious tap of investor capital. Amodei specifically cited the threat of AI-assisted “bioterrorism” in his weekend blog post—following logically from his company’s own internal investigations into suspicious biological research by Claude users, albeit conveniently without ever mentioning Anthropic’s own bio lab.
Of course, pharmaceutical research and development is not the same thing as bioweapon development, but there is so much overlap between the two that processes for one could easily be repurposed for the other, and that includes AI models trained for biological research.
In a lengthy X post this week, MIT synthetic biologist Kevin Esvelt, long a voice of caution in this field, also weighed in. Esvelt, notable for his work advancing how CRISPR could be used to speed-up evolutionary selection via gene drives, revealed that an AI chatbot had recently suggested to him “a novel form of bioweapon” that he “hadn’t realized was possible.”
Esvelt added that he ran his X post by Anthropic’s advanced version of Claude AI, Fable, for factchecking help, “triggering a classifier” in the process. Nice to know, at least, that one safeguard is working.
Read the full article here
