arXiv cs.CL· Zailong Tian·· 2 天前AI 评分42
LLM 会继承另一模型的“指纹”:API 中转审计面临的新挑战
Contagious Fingerprints: When LLMs Inherit Another Model's Identity
AI 导读
研究发现 LLM 的文本指纹具有“传染性”:模型接续另一模型开启的对话后,会继承后者的指纹。在 DeepSeek 替换 Claude Sonnet 的交接后,分类器检测替代模型的准确率从 0.97 降至 0.72,同厂商降级至 Haiku 则从 0.87 降至 0.54。
正文 · 原文
Abstract:The fingerprint by which a classifier recognizes the LLM behind a text is usually treated as a fixed property of the model. We show that it is contagious: a model that continues a conversation begun by another model inherits that model's fingerprint. This undermines audits of API relays, which can serve the first turns of a conversation with the model a customer pays for and silently hand the rest to a cheaper one. After such a handoff, detection of the substitute by a text classifier drops from 0.97 to 0.72 (DeepSeek replacing Claude Sonnet) and from 0.87 to 0.54 for a same-vendor downgrade (Haiku), and retraining on handoff data repairs only substitutes seen in training. Preregistered experiments trace the contagion to imitation of the history's style. History also moves a model away from its own fingerprint: active probes are not camouflaged, but when calibrated on fresh conversations they flag \emph{honest} models on long contexts at rates up to 1.00, and a standard MMD equality test rejects all nine models we test. Both failures come from comparing a history-dependent reply with a history-free reference. We propose \textsc{Lineup}, which samples references from the claimed model \emph{on the same history} and tests the audited reply by its rank among them. \textsc{Lineup} controls false positives by exchangeability and needs no fitted threshold or log-probabilities. With contemporaneous references on two closed models, its false-positive rate is 0.056 (preregistered bound 0.10), and it detects 0.81 of substitute sessions across six pairs after four audited turns.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2611.18104 [cs.CL] |
| (or arXiv:2611.18104v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.18104 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Zailong Tian [view email]
[v1]
Fri, 13 Nov 2026 12:32:44 UTC (169 KB)
来源:arXiv cs.CL · arxiv.org