arXiv：cs.LG· Thiago Sandoval, Ufuk Topcu·· 4 小时前AI 评分46
RCV：面向安全分类器适配与监控的正确性估计方法
Regime-Conditional Verification: Correctness Estimation for Adapting and Monitoring Safety Classifiers
AI 导读
研究者提出 Regime-Conditional Verification（RCV），一种无需重训即可适配现成安全分类器的轻量封装：它从分类器内部表征估计每次预测与部署方策略不一致的概率，并选择性纠正可能错误的预测。
正文
Abstract:Safety classifiers deployed with large language models often fail for two reasons: their decisions reflect the policy learned during training rather than the deployer's desired policy, and their performance degrades as deployment traffic evolves. We present Regime-Conditional Verification (RCV), a lightweight wrapper that adapts an off-the-shelf safety classifier without retraining it. RCV estimates, from the classifier's internal representations, the probability that each prediction disagrees with the deployer's policy, and selectively corrects predictions likely to be wrong. The same correctness estimates also provide a label-free signal for detecting distribution shift, enabling a maintenance loop that updates the correctness estimation layer and resorts to classifier fine-tuning only when repair fails within a label budget. Across three off-the-shelf safety classifiers and two benchmark datasets, RCV improves adherence to the deployer's policy in every classifier-dataset combination, catching up to 0.81 of previously missed unsafe content without modifying the underlying classifier. In a deployment study with ten attack campaigns, each a harm category held out of RCV's training, RCV detects every campaign in a dedicated injection panel; in the maintenance census most drift episodes are repaired without updating the classifier, and the fine-tune is reserved for the residual episodes.
| Comments: | 18 pages including technical appendix, 6 figures. Project page and code: this https URL |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Cryptography and Security (cs.CR); Machine Learning (cs.LG) |
| Cite as: | arXiv:2608.14089 [cs.AI] |
| (or arXiv:2608.14089v3 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2608.14089 arXiv-issued DOI via DataCite |
Submission history
From: Thiago Costantin Sandoval [view email]
[v1]
Fri, 14 Aug 2026 08:50:51 UTC (239 KB)
[v2]
Sun, 20 Sep 2026 10:04:42 UTC (238 KB)
[v3]
Tue, 6 Oct 2026 15:35:38 UTC (241 KB)
来源：arXiv：cs.LG · arxiv.org