Claude found a previously undescribed enzyme system. Anthropic is establishing its own biology laboratory

Anthropic is establishing its own group and laboratory for natural-science research. Its first published result is the discovery of a previously undescribed enzyme system that Claude identified while analyzing biological data with only broadly formulated guidance from scientists. Researchers asked the model to look for uncharacterized protein families in large DNA datasets. Claude found a system with repeating genetic sequences and properties resembling CRISPR. Conventional laboratory work followed to test whether the computational hypothesis matched real biology. The combination of machine search and physical experimentation is more important than the individual discovery itself. Biology contains enormous numbers of sequences, proteins and relationships that humans cannot inspect one by one. An agent can generate and discard thousands of hypotheses in parallel and pass only the most promising ones to scientists. This could change the economics of basic research. Today, even the work required to formulate suitable experiments can be expensive. If AI greatly increases the number of high-quality hypotheses per scientist, laboratory capacity may become the bottleneck: molecule synthesis, cultivation, measurement and other physical experiments. At the same time, this is an important brake on exaggerated expectations. A language model can find an unusual pattern in data, but a text output alone cannot prove that a new biological mechanism actually works. The result must go through experimentation, replication and expert review. Anthropic is entering a field in which Google DeepMind, Isomorphic Labs and numerous biotechnology companies already operate. Strategically, it is notable that a laboratory developing general-purpose models does not want to remain only a tool supplier to other scientists. It is beginning to produce scientific results itself. If this model works, a new type of research organization could emerge. A small number of scientists would direct large volumes of agent computation that search literature, databases and possible experiments. A physical laboratory would then validate the most promising candidates. The business value is particularly clear in pharmaceuticals and biotechnology. The ability to find an interesting molecule or biological mechanism faster can create patentable assets and shorten the early stage of product development. At the same time, questions arise about ownership and reproducibility of results created with substantial AI assistance. Scientific institutions will need to record not only the laboratory procedure but also the model version, its instructions, the databases used and the method of selecting hypotheses. The biggest change, then, is not the statement that “AI discovered an enzyme.” What matters is that a workflow is beginning to form in which an agent actively proposes new directions for scientific research and the human moves into the role of client, critic and experimental verifier. KEY TERMS Enzyme: A protein or other biomolecule that accelerates a particular chemical reaction. CRISPR: A natural bacterial defense mechanism whose components are used for editing genetic information. Research hypothesis: A testable explanation or assumption that can be examined experimentally.

Claude found a previously undescribed enzyme system. Anthropic is establishing its own biology laboratory

The full article is available to users registered at Hard Skills