arXiv cs.AI· Patrikas Vanagas, Augustas Ma\v{c}ijauskas, Laurynas Lopata·· 12 小时前AI 评分31
Capability-Gated Language Models：安全可组合，效用不可组合
Capability-Gated Language Models: Security Composes, Utility Does Not
AI 导读
论文提出“能力门控部署”（capability-gated deployment），在单套模型权重内实现按主体（principal）的访问控制，配置构成格结构，meet 累积限制、join 汇聚联盟可达范围。
正文 · 原文
Abstract:Deployed language model safeguards (safety fine-tuning, filtering, unlearning) vary by principal only outside the model weights: filters are reconfigured, tiers are multiplied, and artefacts are reissued; inside one set of weights every request meets the same model configuration. This motivates us to define capability-gated deployment: per-principal access control inside one set of weights, whose configurations form a lattice - meets accumulate a principal's restrictions and joins pool a coalition's reach. We instantiate it by sparse rank gating over an existing nested-factorisation mechanism, guide profile search with one-pass attribution, and read every result once from a pre-registered held-out split. Security approximately composes: provably exactly at meets under a monotone-elicitation assumption we falsify pointwise. In two lineages the median held-out meet deepens suppression; the one effect surviving correction strengthens it. Utility does not: individually harmless profiles can compose to retention and fluency damage, and no compositional bound exists.
| Comments: | Accepted to NeurIPS 2026 Workshop on Foundations of Language Model Security |
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2609.00445 [cs.CR] |
| (or arXiv:2609.00445v3 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2609.00445 arXiv-issued DOI via DataCite |
Submission history
From: Patrikas Vanagas [view email]
[v1]
Mon, 31 Aug 2026 22:35:49 UTC (379 KB)
[v2]
Mon, 5 Oct 2026 03:08:17 UTC (390 KB)
[v3]
Fri, 16 Oct 2026 03:39:57 UTC (390 KB)
来源：arXiv cs.AI · arxiv.org