One integer sequence can identify language models with high accuracy
One Sequence Is Enough: Fingerprinting Language Models from Fixed-Length Integer Sequences
Computation and Language
Summary
Sometimes people use fake names to pretend they are a certain AI model. The researchers found that by asking a model to generate a specific sequence of numbers in one go, they can tell which model it really is just from that sequence. They made a test set of many such number responses from 51 models and developed a smart program that guesses the model with over 92% accuracy from just one answer. Using more answers makes this identification even better. This helps verify a model’s identity based only on its output.
What this means in practice
- •For api service providers: Verify the true identity of AI models behind API endpoints by analyzing fixed-length integer sequence outputs.
- •For platform operators: Detect mismatch between declared and actual AI models powering third-party services by fingerprinting their numerical responses.
Authors
Fantong Meng, Chengjie Sun, Shuangjia Yang, Zhiyuan Zhao, Yansheng Pei, Xinyang Li
Abstract
Users of model APIs and third-party platforms often rely on the declared model name to identify the model responding to their requests. However, the actual model may differ from the declared one, motivating identity verification based on observable outputs. Previous studies have shown that preferences in simple numerical-choice tasks can serve as model fingerprints. We investigate whether a single fixed-length integer sequence can combine multiple numerical observations to identify its source model. For this task, we introduce NumPrint, a benchmark of 5,036 answers from 51 models, collected using Chinese, Japanese, and English prompts that request 294 integers between 1 and 355 in one response. We propose the Shallow MLP and Causal Transformer (SMCT) to capture aggregate numerical preferences and behavioral patterns in generated sequences. In 51-class closed-set evaluation, SMCT achieves 92.72% accuracy from a single sequence, outperforming all methods compared in this study. Using two and four answers increases accuracy to 96.06% and 97.12%, respectively. These results show that a single integer sequence contains useful model fingerprints and that additional answers further improve identification.