NumPrint identifies models from one integer sequence in a 51-model test
Text
A preprint explores numerical fingerprints behind model responses.
A string of integers can reveal which AI generated it. NumPrint's researchers report 92.72% identification accuracy from one response among 51 known models.
The November 10 preprint uses numerical preferences as fingerprints. The motivation is practical: a third-party service's model label does not itself verify its identity.
This is a closed-set experiment, not authentication for every anonymous API. Updated or previously unseen models would need separate evaluation.
The work suggests another auditing direction: examine behavior alongside platform disclosures. One classification should not become an accusation of fraud.
Source published: arXiv, November 10, 2026, 20:50:33 Beijing / 07:50:33 EST.
Share this article
What do you think of this story?
Anonymous feedback, up to 5 reactions per article.
Anonymous comments
Share your thoughts without signing in.
Discuss the technology and the story. Hate, politics and war, threats, abuse, personal accusations, sexual content, ads, links and personal information are not allowed.
No comments yet. Be the first to share.