Jie Wang (Southwest University of Political Science and Law) has posted Proof After Compression: Copyright Infringement in AI Systems on SSRN. Here is the abstract:
Copyright proof has long established copying through a connection between a specific source and a specific output. When millions of works are compressed into a single model, the same process that gives the model generative power eliminates any record of which source contributed which feature of which output. That information is gone. Access and similarity remain observable, but what they once revealed is no longer recoverable at the level of detail prior doctrine assumed. The conditions that any such connection would have to satisfy, however, remain within the law’s reach. Three are observable: the circumstances of training, the model’s capacity to retain what it absorbed, and the features of the output itself. The fourth is settled only by elimination: whether any explanation independent of the source accounts for the resemblance. Copyright proof must shift from reconstructing the connection to verifying those conditions. This is proof after compression. Proof after compression works within copyright’s existing commitments. Strict liability, the idea-expression distinction, plaintiff-side proof of copying, and adversarial adjudication all remain in place. Verification alone would let source-specificity collapse into protectability. The exclusion method holds the two apart: only a source-specific residuum that survives the elimination reaches copyright’s unchanged protectability inquiry. Condition verification must also preserve rational inference, respect doctrinal limits, and permit meaningful adversarial contestation: three requirements this Article institutionalizes as Risk Path Proof, a four-element evidentiary framework that allocates burdens by observability and structures the defendant’s challenge through specific alternative explanations. The leading AI copyright cases show what the framework must do and where its limits fall. The failure they expose has a doctrinal answer. The diagnosis beneath the failure reaches further.
Recommended!
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum