When AI Is Accused of Lying, Who Lies? Error, Constraint Arbitration and the AI-Mediated Lie

Artificial intelligence (AI) systems are increasingly accused of lying, and users are offered prompts said to force them to tell the truth. A significant class of these accusations conflates distinct phenomena. At output level, this note separates a functional lie, a mere error, signalled constraint arbitration and unsignalled constraint arbitration. A false answer may faithfully reflect a false, incomplete or biased representation; a selective answer may result from competing requirements rather than from insincerity. The central practical failure arises when a limited analysis is presented as complete or better supported than it is. The note then asks a different question: who lies? It distinguishes three attributional configurations. A system may lie in a functional sense when it asserts what it represents as false so that it will be believed. A human or organisational actor may also lie through the system. This note proposes the term AI-mediated lie for the strict lie attributable to an identifiable actor who intentionally uses, configures, supplies or manipulates an AI system so that it communicates as true what that actor takes to be false, intending the recipient to believe it. The system may generate, personalise, amplify or transmit the assertion and may itself remain honest relative to a deliberately manipulated reference environment. Finally, a chain may deceive without any identifiable liar when qualifications are progressively lost. The phenomena of AI-mediated deception, malicious use, corpus poisoning and adversarial environmental manipulation are established. The proposed contribution is narrower: an attributional category separating the actor who lies from the AI that mediates the lie, and distinguishing both from structural deception. The note proposes three governance requirements: protect and govern sources and reference environments; declare and govern priority ordering; and signal constraint arbitration, preserving its signal through delegated hand-offs. It also distinguishes causal contribution, responsibility and legal liability. The note concludes by marking the transition to a forthcoming comparative study of European Union, Chinese and United States law.