Publishers have a new trick: they can hide invisible 'traps' in their work that make it legally impossible for AI to learn from them.
While most legal battles against AI focus on copyright, this paper identifies a technical 'trap' where the automated cleaning of HTML—a necessary step for AI training—forces the AI to intentionally remove licensing metadata. This structural 'entanglement' transforms a simple web-crawling task into a violation of the DMCA.
Compound Statutory Liability Entrapment in Inference-Time AI Pipelines
SSRN · 6432898
As Artificial Intelligence shifts from static training ingestion to real-time, inferencebased retrieval, the economic harm to primary web publishers has accelerated. Current legal frameworks focus heavily on 17 U.S.C. § 501 (infringement of the exclusive rights granted under § 106), which is frequently obfuscated by "Fair Use" defenses. That debate is already old. The industry has moved on. This paper introduces a novel enforcement paradigm utilizing 17 U.S.C. § 1202 (Integrity of Copyright Mana