Criminal Law Regulation of Network Rumors Generated by Autonomous Operation of Artificial Intelligence Programs

Abstract

When large language models generate text without human oversight-a capability that generative artificial intelligence has now made routine-the question of who bears criminal law responsibility for harmful output becomes acute. Online misinformation produced by autonomous programs does not fit neatly into doctrines built on the assumption of a human perpetrator. This article tackles that mismatch. It begins by unpacking the technical roots of AI-generated rumors: flaws embedded in software architecture, the volatile behavior of self-learning algorithms, and situational triggers in the deployment environment. These factors, separately and in combination, explain why fully autonomous rumor generation is not merely theoretical. The legal fallout is threefold, though not neatly compartmentalized. Pinpointing a responsible subject is elusive when the "actor" is code; subjective intent dissolves where there is no mind to read; and the causal chain from algorithm to social harm disappears into opacity. Addressing these failures, the study argues for a "program legal liability" framework predicated on "breach of program security obligations"-a standard that casts a duty net over developers, operators, and users alike. Alongside it runs a dual-track accountability structure: "actor liability" for human participants and "program liability" for the autonomous system itself. The argument culminates in "algorithm risk creation," a concept that leverages objective imputation theory to bypass the evidentiary dead-end of the "black box," drawing a defensible line between criminal and non-criminal outcomes. Together these elements reframe how criminal law should confront the challenge of self-generating misinformation.