AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

Auditor-Assisted Summary-Channel Verification for Hosted LLM Identity Substitution

Ziyi Zhang, Ziyao Wang, Guoheng Sun, Ang Li, Jian Li

NeurIPS 2026 · Large language models › Reasoning and chain of thought · approach: Routing, pruning and distillation for reasoning · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes an auditor-assisted regime for verifying whether a hosted LLM service serves the claimed model identity, using pre-enrolled verification material and audit-time checks on released outputs only.

Problem

Hosted LLM providers may claim a model identity that differs from the configuration actually serving requests, and external users and auditors cannot inspect the served model at runtime.

Approach

A trusted auditor enrolls claimed identities and prepares verification material before deployment; audit-time verification then uses only the claimed identity and the released output, with no runtime access to the service.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use identity-specific keyed carrier sets with adapters instead of generic watermarks (Unigram, Gumbel-Max, SynthID) (ablation)SumMark · Math substitution detection, 7 pairsTPR@1% FPR: Unigram 0.4730 -> 1.0000 on Qwen-7B to Qwen-3BImproved
summary-channel watermark instead of unkeyed text classifiersSumMark · General substitution detectionaverage TPR@1% FPR: 0.5439 -> 0.9750 (vs BERT)Improved
summary watermarking on task-facing answers (ablation)SumMark · Math and Codeanswer accuracy / Pass@1: Qwen-7B Math 0.5034 -> 0.5062; Code 0.1920 -> 0.1830No clear effect
summary watermarking on General summaries (ablation)SumMark · Generalperplexity: Qwen-1.5B 29.3089 -> 25.6346Improved
bounded edits (shuffle, deletion, truncation) to released summary (ablation)SumMark · Math robustnessTPR@1% FPR: 0.9994 -> 0.9992No clear effect
bilingual rewriting of summary (ablation)SumMark · Math robustnessTPR@1% FPR: 0.9994 -> 0.7900 (worst 0.5740)Hurt
adapter distillation spoofing attack (ablation)SumMark · Math robustnessASR@1% FPR: average 0.1651, worst 0.6290Hurt
watermark stealing spoofing attack (ablation)SumMark · Math robustnessASR@1% FPR: average 0.0417No clear effect
quantize served model from FP16 to INT8 (small model) (ablation)SumMark precision diagnostic · Qwen-1.5B MathTPR@1% FPR: INT8 0.1132 vs INT4 0.3417Hurt
more aggressive quantization (INT4 vs INT8) (ablation)SumMark precision diagnostic · Qwen-1.5B, MathTPR@1% FPR: 0.1132 -> 0.3417Improved
use SumMark instead of T5 text classifier on templated Code summariesSumMark attribution · Code attributionaccuracy: T5 0.9841 vs SumMark 0.9678Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy