PHRONESIS: Systemic Friction Hypotheses — Final v1.0

This paper examines whether modeled systemic friction is sufficient evidence that an adapting optimizer will settle on restraint. Its principal exact result is a synthetic counterexample to the inference that fixed-policy self-penalty is sufficient evidence for natural restraint. In the stipulated model, constant maximal manipulation becomes self-penalizing while an adaptive manipulate-and-recover policy remains profitable, and an optimal infinite-horizon policy does not converge to zero manipulation under the stated assumptions. The result is therefore a counterexample to a specific inference pattern within the model, not a general claim that adaptive optimizers resist restraint or that systemic friction cannot support cooperation. Physical-exposure and deception-maintenance mechanisms remain empirical conjectures, and the proposed empirical program may be ethically or operationally untestable in some high-risk deployments. The paper is descriptive and non-authoritative; independent specialist peer review has not been obtained and empirical validation has not been performed. For categories, I recommend starting with Artificial Intelligence Safety as the primary category. PhilPapers says works should normally be assigned to only one or two genuinely central leaf-level topics, rather than every category they touch. Artificial Intelligence Safety is an established PhilPapers category under Philosophy of Artificial Intelligence. A reasonable second category is Ethics of Artificial Intelligence, Miscellaneous if the interface allows it and you want the autonomy/restraint implications represented. I would not use Machine Ethics unless you want to claim that ethical-machine design is a principal focus, because PhilPapers describes that category more narrowly.