Version2: Why Stochastic-AI Regulation Is a Fallacy And Yet, Planetarily Discussed

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The global debate over artificial intelligence regulation rests on a premise that has neverbeen examined: that the object of regulation—the stochastic large language model—is agovernable entity. This paper argues that it is not, and that the entire planetary discussion is therefore a fallacy in the precise sense: not a mistake of degree, but a category error of kind. "Stochastic" here does not mean merely sampling with temperature greater than zero. It means the paradigm of unbounded, opaque, non-stationary learned functions. Deterministic decoding does not escape the argument, because the impossibility is structural in the specification, not in the sampler. The argument proceeds in five parts. First, it distinguishes five kinds of standard that the word "standard" conceals, only one of which—functional safety—certifies behavior, and shows that this one is theoretically impossible for the stochastic paradigm. Second, it demonstrates that the impossibility is architectural, not incremental: the paradigm's defining property is the absence of a total behavioral specification, and hallucination is a mathematical consequence of calibration, not of sampling. Third, it shows that the four existing functional safety standards—ISO 26262, IEC 61508, DO-178C, IEC 62443—already exclude this paradigm from the safety path, and that ISO/IEC TR 5469:2024 exists specifically to describe this gap without closing it. Fourth, it explains why the debatecontinues anyway: incentive capture, institutional vacancy, and the performative function of governance certification. Fifth, it states what follows. The fallacy is not that regulation fails. The fallacy is that the object of regulation cannot be governed, and that a deterministic alternative—already specified, already open, already deployable—has been excluded from the conversation by the same forces that sustain the debate. The paper concludes that the correct response is not better regulation of the ungovernable,but adoption of the standard that governs the layer where governance is technically possible. Keywords: AI regulation, functional safety, certification impossibility, regulatory capture,category error, deterministic architecture, specification