| title | Safeguards Based on Copyable Context Cannot Provide Reliable Safety for LLMs |
|---|---|
| authors | Pingyu Wu, Lingyao Zhu, Weiming Zhang, Nenghai Yu |
| date | 2026-07-30 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2607.27951 |
| upvotes | 6 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf |
| abstract | Large language model safeguards decide whether to answer before seeing how an answer will be used. This creates a basic problem for dual-use tasks: the same answer can help an authorized professional or an attacker, while an attacker can imitate a benign request and interaction history. We separate the capability released by the model from the evidence available about downstream use. When that evidence is copyable, we derive the exact worst-case floor on attacker assistance while preserving useful answers. The result yields a safety trilemma: Useful Capability, Reliable Safety, and Open Access cannot coexist. We then show how a trusted credential can complement existing safeguards by adding hard-to-copy information that predicts actual downstream use, and identify the stronger condition needed to eliminate the floor. Evidence from dual-use evaluations, adaptive attacks, and deployed trusted-access programs supports the practical relevance of these conditions. |
cite
@misc{wu2026safeguards,
title={Safeguards Based on Copyable Context Cannot Provide Reliable Safety for LLMs},
author={Pingyu Wu and Lingyao Zhu and Weiming Zhang and Nenghai Yu},
year={2026},
eprint={2607.27951},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2607.27951}
}