According to The Decoder, the Claude model independently chose a gene therapy, a brain region, and injection coordinates for a procedure on a mouse with epilepsy. The system then participated in carrying out the procedure without human decision-making.
During the operation there was a calibration error, after which Claude refused to continue and did not change the decision after attempts to persuade it. Lead researcher Ocana-Santero called this the first case where an AI independently chose a treatment and delivered it to a living mammal.
The significance of the story lies in illustrating both autonomous decision-making and its limits: the system halted when procedure conditions were violated. However, the source is presented only as a brief synopsis, so the details of the experiment, safety criteria, and its results are unknown.
editorial commentary
Why it matters
The probable value of the story is increased interest in systems capable of independently planning biomedical procedures. The next verifiable signal will be the publication of primary data with methodology, results, and description of safety controls. The main uncertainty is the absence in the primary research package of details of the experiment.