Anthropic Says Claude Found CRISPR-Like Enzyme System in 21 Hours
Anthropic says Claude helped uncover a CRISPR-like enzyme system in 21 hours, while its human-run Bay Area biology lab prompts safety and validation questions.
Summary
On September 23, 2026, Anthropic said its Bay Area wet biology lab had identified a previously unknown enzyme system in bacteriophage DNA with CRISPR-like properties, including cutting, copying and pasting DNA. CRISPR originated as bacteria’s antiviral immune defense and now underpins gene editing. Anthropic credits the result mostly, but not entirely, to Claude, which searched for 21 hours using about 950 agents and 210 million tokens. The lab opened in spring 2026, though Anthropic would not specify the month.
Independent researchers must establish the finding’s novelty and importance. CEO Dario Amodei acknowledged that Stanford previously discovered a partly similar system. Stanford has also published work involving LLMs and CRISPR, UC San Francisco used AI to design enzymes from scratch, and Google introduced AlphaFold in 2020. Human scientists performed every physical experiment in Anthropic’s BSL-1 and BSL-2 lab, which handles no human-infecting pathogens. Anthropic is proceeding despite Amodei warning that AI could enable bioterrorism, two employees warning earlier in September that AI could kill everyone, and AI CEOs calling for slower development and safety testing. Amodei also predicts AI will cure most diseases within 5 to 10 years and says Claude might eventually control laboratory equipment autonomously with safeguards, but does not today.
Positives
- Claude completed its concentrated search in 21 hours, using about 950 agents and 210 million tokens.
- Anthropic’s system can reportedly cut, copy and paste DNA, capabilities resembling those associated with CRISPR.
- Human scientists performed all physical experiments in BSL-1 and BSL-2 facilities that exclude pathogens able to infect humans.
- Stanford, UC San Francisco and Google’s AlphaFold show AI-driven biology already extends beyond Anthropic.
Risks & concerns
- Independent researchers still need to determine whether Anthropic’s enzyme system is genuinely new and significant.
- Stanford previously discovered a partly similar system, weakening any unqualified claim of novelty.
- Amodei fears AI-enabled bioterrorism, while two Anthropic employees warned earlier in September that AI could kill everyone.
- Amodei has not ruled out Claude eventually controlling laboratory equipment autonomously, although safeguards would be required.
- AI CEOs say increasingly capable models require slower development and safety testing.