aiqbench analysis
This research exposes a critical vulnerability in the anonymity of large language model leaderboards, demonstrating how deep stylistic patterns can be extracted via interpolated preference data to de-anonymize models. By synthesizing hard negative samples and employing adaptive curriculum learning, the proposed INTERPOL framework significantly outperforms superficial statistical baselines in identifying specific model identities. The study further quantifies the real-world threat of these findings through simulations showing how such identification capabilities could manipulate ranking systems on platforms like LM Arena.
InterPol: De-anonymizing LM Arena via Interpolated Preference Learning
arXiv:2603.15220v2 Announce Type: replace Abstract: Strict anonymity of model responses is a key for the reliability of voting-based leaderboards, such as LM Arena. While prior studies have attempted to compromise this assumption using simple statistical features like TF-IDF or bag-ofwords, these methods often lack the discriminative power to distinguish between stylistically similar or within-family models. To overcome these limitations and expose the severity of vulnerability, we introduce INTERPOL, a model-driven identification framework that learns to distinguish target models from others using interpolated preference data. Specifically, INTERPOL captures deep stylistic patterns that superficial statistical features miss by synthesizing hard negative samples through model interpolation and employing an adaptive curriculum learning strategy. Extensive experiments demonstrate that INTERPOL significantly outperforms existing baselines in identification accuracy. Furthermore, we quantify the real-world threat of our findings through ranking manipulation simulations on Arena battle data.
Abstract:Strict anonymity of model responses is a key for the reliability of voting-based leaderboards, such as LM Arena. While prior studies have attempted to compromise this assumption using simple statistical features like TF-IDF or bag-ofwords, these methods often lack the discriminative power to distinguish between stylistically similar or within-family models. To overcome these limitations and expose the severity of vulnerability, we introduce INTERPOL, a model-driven identification framework that learns to distinguish target models from others using interpolated preference data. Specifically, INTERPOL captures deep stylistic patterns that superficial statistical features miss by synthesizing hard negative samples through model interpolation and employing an adaptive curriculum learning strategy. Extensive experiments demonstrate that INTERPOL significantly outperforms existing baselines in identification accuracy. Furthermore, we quantify the real-world threat of our findings through ranking manipulation simulations on Arena battle data.
Artificial Intelligence (cs.AI)
arXiv:2603.15220 [cs.AI]
(or arXiv:2603.15220v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2603.15220 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 226 words
Related coverage
How to apply to AI safety fellowships (and beyond)
LessWrong · Oct 2, 2026
LitReview Arena: Evaluating Literature Review Agents with Battle-Style Peer Review Platform
arXiv · Aug 25, 2026
Current AIs out-persuade professionals in lab settings but (probably) not in the world
LessWrong · Oct 5, 2026
[AINews] not much happened today
Latent Space · Oct 3, 2026