A random number can identify an AI model: behavioral fingerprinting can detect API relay model swapping
Original title:AI说不出的随机数,成了鉴别套壳大模型最好的照妖镜。
AI summary
Tomas Bruckner, a researcher at the Prague University of Economics and Business, found that repeatedly asking a model to output a random number from 1 to 100 can produce a unique "behavioral fingerprint." After asking 165 models 30 times each, he found GPT-4o favored 42 and 37, Claude Sonnet 5 overwhelmingly output 47, and Qwen3-Max answered 42 all 30 times. The method needs only about 120 requests to identify a model, with an error rate of about 10.6%, offering a lightweight way to verify whether an API has swapped models.
Key Takeaways
- Tomas Bruckner,a researcher at the Prague University of Economics and Business,found that…
- " After asking 165 models 30 times each,he found GPT-4o favored 42 and 37,Claude Sonnet 5…
- The method needs only about 120 requests to identify a model,with an error rate of about 10.
- 6%,offering a lightweight way to verify whether an API has swapped models.
Quick Facts
- Published
- Updated
- Source
- 公众号:数字生命卡兹克
- Category
- Tutorials & Tips
- AI score
- 64
一个Token识别模型的妙招,轻巧但准确率高,对于被API中转站坑过的人来说简直是照妖镜,而且读起来像在破案。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: A random number can identify an AI model: behavioral fingerprinting can detect API relay model swapping. July 21, 2026. https://new.nat6.net/en/news/cmrtyvkng3voibihzero6pry1/