Position: No Retroactive Cure for Infringement during Training

TL;DR

This paper argues that post-hoc mitigation methods like unlearning cannot retroactively address legal liability in AI training, emphasizing the importance of verifiable ex-ante compliance.

Contribution

It presents a legal and technical argument for shifting from post-hoc mitigation to verifiable ex-ante process compliance in AI training.

Findings

Post-hoc mitigation cannot undo unlawful data acquisition.

Legal restrictions can bypass traditional copyright defenses.

Model weights may retain value from protected inputs, affecting remedies.

Abstract

As generative AI faces intensifying legal challenges, the machine learning community has increasingly relied on post-hoc mitigation -- especially machine unlearning and inference-time guardrails -- to argue for compliance. This paper argues that such post-hoc mitigation methods cannot retroactively cure liability from unlawful acquisition and training, because compliance hinges on data lineage, not the outputs. Our argument has three parts. First, unauthorized copying/ingestion can be a legally complete completed act, and model weights may operate as fixed copies that retain training-derived expressive value, making later filtering beside the point for infringement. Second, contract and tort/unfair-competition rules -- via licenses, terms of service, and anti-free-riding principles -- can independently restrict access and use, often bypassing copyright defenses (e.g., fair use or TDM exceptions). Third, since value from protected inputs can persist in weights, remedies such as unjust enrichment and disgorgement may require stripping gains and, in some cases, reaching the model itself. We therefore argue for a shift from Post-Hoc Sanitization to verifiable Ex-Ante Process Compliance.