Since 2023 a handful of research groups have published systems that read the literature, propose hypotheses, write and run code, and in some cases write the paper as well. Their authors call them AI scientists, and the cost figures they report have made a new idea thinkable: a laboratory of one person, in which a single researcher directs software that does most of the work a group once did. This paper asks, in Meridian's form, when a result produced by such a system may stand as a scientific finding, who decides, by when, and what would make the decider revisit the decision. It answers from ten public texts and nothing else: seven papers that present six such systems, a perspective on their risks, a United States executive order that assigns a federal programme of AI-driven science, and the national AI action plan that preceded it. It does not use IATHEON's records, and it gives no medical, regulatory or investment advice.
The texts never let a system decide. Each one, read closely, places the decision with people at named points: the scientist who sets the objective and prepares the data, the person who selects among outputs, the reviewer, the venue, and the authors who sign. The deadlines the texts set bind institutions, not findings. The conditions for looking again are written as failures the authors themselves found: invented results, contaminated tests, runs that do not repeat, and claims stronger than their evidence. A laboratory of one gathers every one of these human points into a single person. The paper's contribution is to show where each text puts that person, and what each leaves for someone else to decide.