Anthropic enters molecular biology research: 950 agents discover an enzyme system candidate for genetic engineering

Anthropic announced its first scientific discovery emerging from its new molecular biology lab in the San Francisco Bay Area, consisting of the detection of a previously unknown enzyme system called ART. The discovery, reached by the AI model Claude, is presented by the company as practical evidence of AI's ability to accelerate basic research and shorten the initial exploration phases in biosciences.

The achievement relied on running roughly 950 software agents supported by the Claude model to comb through a genetic sequence database continuously for 21 hours.The software team scanned more than 200,000 genes of a single enzyme type and shortlisted 20 genes as candidates for precise laboratory examination.A task that normally takes weeks or even months of specialist researchers' work to produce a list of experimentally testable candidates.

The breakthrough arose when an agent identified a genetic structure in bacteriophages that had previously escaped scientists' attention, consisting of a series of short, repeated DNA fragments flanking two genes: one for a known enzyme and another for a unknown protein. Preliminary experiments suggest that the ART system may be programmable in a manner similar to CRISPR, prompting geneticist Feng Chang of the Massachusetts Institute of Technology and the Broad Institute to describe the observation of these repeats as interesting and worthy of investigative follow-up. The results were published in a preliminary research paper that has not yet undergone peer review, coinciding with Darío Amodei's note that a team at Stanford University independently observed a similar yet distinct system.

The chief executive of Anthropic, Darío Amodei, chose a United Nations Security Council session, addressing it via a televised segment, to present the discovery as a preliminary result that could lay the groundwork for a new mechanism in genetic editing serving future gene therapies. Human researchers' role was limited to drafting initial directives and conducting laboratory experiments, while the agents operated independently to select promising pathways. The lab, founded in spring of this year, also emphasized that it does not handle pathogens capable of infecting humans, that all laboratory work is performed by human researchers, and Amodei did not rule out the possibility of Claude operating laboratory equipment in the future once adequate safeguards and controls are in place.

This shift in biomedical research approach is reshuffling priorities at research institutes and startups in the Gulf and Egypt, where national genome projects and biotech investments top health sector development plans. What changes in practice for laboratory managers and biotech teams in the region is the bottleneck moving from the information-mining andThis shift in biomedical research approach is reshuffling priorities at research institutes and startups in the Gulf and Egypt, where national genome projects and biotech investments top health sector development plans. What changes in practice for laboratory managers and biotech teams in the region is the bottleneck moving from the information-mining and big-data analysis stage, software agent swarms have shown they can complete it in a day with modest computing cost, to the wet-lab verification stage. This means the competitive edge no longer depends on the number of staff assigned to manual biological sorting, but on the speed of linking automated testing agents with field labs capable of isolating samples and biologically validating them.

The skill demanded in the regional work environment is clearly moving toward biologists capable of crafting precise instructions for AI agents and auditing their genetic hypotheses, while maintaining the highest laboratory safety standards before transferring outputs to therapeutic applications.