SCOPE: Intrinsic Semantic Space Control for Mitigating Copyright Infringement in LLMs

TL;DR

SCOPE is an inference-time method for large language models that mitigates copyright infringement by controlling semantic space representations without needing parameter updates or external filters.

Contribution

It introduces a novel semantic-space control approach using sparse autoencoders to identify and clamp copyright-sensitive subspaces during inference.

Findings

Mitigates copyright infringement effectively

Maintains model utility and performance

Provides interpretability of semantic subspaces

Abstract

Large language models sometimes inadvertently reproduce passages that are copyrighted, exposing downstream applications to legal risk. Most existing studies for inference-time defences focus on surface-level token matching and rely on external blocklists or filters, which add deployment complexity and may overlook semantically paraphrased leakage. In this work, we reframe copyright infringement mitigation as intrinsic semantic-space control and introduce SCOPE, an inference-time method that requires no parameter updates or auxiliary filters. Specifically, the sparse autoencoder (SAE) projects hidden states into a high-dimensional, near-monosemantic space; benefiting from this representation, we identify a copyright-sensitive subspace and clamp its activations during decoding. Experiments on widely recognized benchmarks show that SCOPE mitigates copyright infringement without degrading general utility. Further interpretability analyses confirm that the isolated subspace captures high-level semantics.