ABSTRACT
Large language models (LLMs) increasingly mediate the drafting, refinement, and stabilization of creative and inventive expression. Intellectual property doctrine traditionally assumes that the boundary between unprotected ideas and protected expression is determined ex post through judicial adjudication. Contemporary LLM systems, however, modulate expressive development ex ante through probabilistic decoding, similarity mitigation, and purpose-sensitive constraint activation. As a result, the crystallization of potentially protectable expression is often conditioned within privately governed computational infrastructures before courts or agencies can review the resulting output. This Article argues that LLM architecture functions as a de facto boundarygovernance mechanism within intellectual property law. By influencing when expression becomes sufficiently determinate, enabling, or repeatable to satisfy doctrinal thresholds, generative systems shape authorship formation, disclosure practices, evidentiary positioning, and innovation incentives across copyright and patent regimes. The principal concern is not intentional appropriation but institutional displacement: expressive opportunity and disclosure standards are recalibrated within opaque safety and risk-optimization frameworks that operate outside traditional public-law oversight.
Dilmaghani, Maryam, The Magnificent Beast: LLM Architecture as Ex Ante Boundary Governance of Intellectual Properties (February 21, 2026).
Leave a Reply