Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2511.07001 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866918195729793024 |
|---|---|
| author | Zhang, Zhenliang Hu, Xinyu Wan, Xiaojun |
| author_facet | Zhang, Zhenliang Hu, Xinyu Wan, Xiaojun |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2511_07001 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | SCOPE: Intrinsic Semantic Space Control for Mitigating Copyright Infringement in LLMs Zhang, Zhenliang Hu, Xinyu Wan, Xiaojun Computation and Language 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. |
| title | SCOPE: Intrinsic Semantic Space Control for Mitigating Copyright Infringement in LLMs |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2511.07001 |