Machine action selection creates neither an automatic responsibility vacuum nor automatic human liability. This paper develops a boundary-accountability framework that separates selection autonomy from authority over effects; declared admissibility from operational reachability; and causal contribution, answerability, fault, and legal liability. The technical core is the declared-effective boundary gap: modeled execution histories reachable under a configuration and threat model but excluded by the deployment's declared scope. A nonempty gap is evidence of a technical mismatch within the model. It does not establish probability, notice, duty, fault, intent, causation, or liability. An empty gap is equally bounded: it can enforce an illegitimate policy, omit real effect paths, or depend on unsound abstraction. A conditional mediation proposition shows that prohibited modeled effects are excluded if every relevant effect passes through an accurate, non-bypassable gate whose allowed transitions preserve the declared safety property. The proposition is a specification theorem, not evidence that a deployment satisfies its premises. Real use requires exact configurations, concretization between abstract and physical effects, current authority, effect observation, bypass testing, custody, and independent verification. The source reports an exhaustive 625-string finite illustration and says that a companion checker is supplied. No such program or result file was found in the authorized corpus. An independent AI reconstruction from the prose reproduced the reported counts, but that is conditional arithmetic validation rather than custody of the claimed author evidence. The source's 2026 incident and legal claims also remain author-supplied pending separately authorized primary-source verification. This revision integrates the accountability framework with Authority Laundering Theory, constitutional enforcement, receipts, observability, work custody, Warrant by Structure, the universal GSM, and the realization funnel. It grants no factual, legal, safety, implementation, or publication standing.
No takes yet. Share an insight, caveat, or question.
Adam Ableman (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: