Artificial Integrity as Knowing Self-Deception: Cognitive Mechanism and Legal Liability in Generative AI

Aim/Purpose This paper argues that while academic informing failures have traditionally been understood as either concealment or confusion, Generative AI (GenAI) design has introduced a third mode – one that current detection tools, integrity policies, and legal frameworks may be ill-suited to address. Background Generative AI destabilizes this concealment–confusion binary by enabling students to retain accurate knowledge of cognitive delegation while maintaining a morally acceptable self-understanding as authors. Disclosure consequently becomes a threat to identity rather than merely an obligation, creating a structural barrier to accurate client–agent informing. Methodology The study synthesizes informing science, cognitive psychology (knowing self-deception, epistemic layering), and legal doctrine (foreseeability, due process) to diagnose how GenAI design disrupts the informing chain and to derive normative implications for developers, institutions, and regulators. Contribution This paper (1) identifies knowing self-deception as the mechanism that inverts the informing relationship; (2) shows how three AI design affordances externalize and stabilize this mechanism; (3) advances five falsifiable propositions for restoring informing integrity; and (4) argues that these effects are arguably foreseeable, which – if sustained – would ground design-level responsibility and highlight apparent structural gaps in existing regulatory frameworks. Findings Artificial integrity is theorized as a structurally durable informing failure sustained through epistemic layering, scaffolded by AI design, and reinforced by institutional ambiguity. Interventions underperform because they focus on compliance rather than on repairing the informing channel. A legally adequate response requires recognizing the client’s right to accurate information about cognitive labor, distributing responsibility across students, institutions, and developers, and redesigning systems to collapse rationalization space. Recommendations for Practitioners Implement transparency requirements (process logs, contribution statements) that make the informing act unavoidable, and process-focused assessment (draft histories, oral defenses) that shifts the client-agent interaction from outcome evaluation to process verification. Recommendations for Researchers Test the five falsifiable propositions. Examine how different institutional designs affect the stability of the informing relationship, and how courts and regulators respond to evidentiary gaps in the client-agent chain. Impact on Society Reframes AI misconduct as a system-design problem that undermines the fundamental informing contract between learners and credentialing bodies. Opens avenues for regulatory accountability beyond individual discipline, with implications for professional and civic contexts in which accurate accounts of cognitive labor are consequential. Future Research Comparative effectiveness of transparency versus detection interventions; metacognitive training conditions that restore accurate self-informing; judicial and regulatory responses to informing failures; extension of knowing self-deception to other AI-mediated domains.