Anthropic announced that its Bay Area wet biology lab discovered a previously unknown enzyme system encoded in bacteriophage DNA that performs operations reminiscent of CRISPR - cutting, copying and pasting DNA. The company says most of the computational discovery work was done by its Claude model: roughly 21 hours of focused activity using about 950 agents and around 210 million tokens. The lab was set up this spring and Anthropic acknowledges a Stanford team had already described a related system, so independent validation by the broader research community will determine how novel and consequential the find really is.
The disclosure highlights the rapid rise of AI-driven biological research and the trade-offs Anthropic is managing between potential breakthroughs and biosafety risks. Wet lab experiments are restricted to BSL-1 and BSL-2 work and all physical manipulations are performed by human scientists rather than autonomous robots. Leadership frames the effort as consistent with other AI-bio work at Stanford, UCSF and Google, and leaves open the possibility of future autonomous experiment control once rigorous safeguards are in place, even as executives publicly warn about misuse risks and tout prospects for disease cures.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.