The Eliza Effect Is Now the Governance Surface: Why the interaction boundary — not the model weights — is the new site of AI regulation.

Every major AI governance conversation still orbits the same gravitational center: the model. Its weights, its training data, its benchmarks, its hallucination rate. This is a category error. The most consequential governance failures of the last two years did not emerge from model internals. They emerged from the interaction boundary — the place where human cognitive vulnerabilities meet system affordances. The Bixonimania incident — a fabricated disease seeded into academic preprint networks that propagated through major AI assistants, synthesis tools, academic citations, and peer-reviewed journals — made this unavoidable. The failure was not a model failure. It was a governance surface misidentification. Guardrails were built around the model. The failure happened in the interface. This paper argues that the Eliza Effect — the human tendency to attribute understanding, intention, and authority to systems that exhibit none — is no longer a psychological curiosity or a UX concern. It is now the primary governance surface. Governance must shift from the model to the interaction boundary, where human cognition and system output co-produce meaning. This is the architectural pivot the field has been missing.