English

Don't Lie to Me: Avoiding Malicious Explanations with STEALTH

Software Engineering 2023-01-26 v1 Artificial Intelligence Cryptography and Security

Abstract

STEALTH is a method for using some AI-generated model, without suffering from malicious attacks (i.e. lying) or associated unfairness issues. After recursively bi-clustering the data, STEALTH system asks the AI model a limited number of queries about class labels. STEALTH asks so few queries (1 per data cluster) that malicious algorithms (a) cannot detect its operation, nor (b) know when to lie.

Keywords

Cite

@article{arxiv.2301.10407,
  title  = {Don't Lie to Me: Avoiding Malicious Explanations with STEALTH},
  author = {Lauren Alvarez and Tim Menzies},
  journal= {arXiv preprint arXiv:2301.10407},
  year   = {2023}
}

Comments

6 pages, 6 Tables, 3 figures

R2 v1 2026-06-28T08:19:23.042Z