Kai Yao

Education

University of Edinburgh

Ph.D. in Cyber Security, Privacy, and Trust

Johns Hopkins University

M.Sc. in Mechanical Engineering

Fudan University

B.Sc. in Theoretical & Applied Mechanics

News

• Dec 2025
  Smudged Fingerprints accepted to the 4th IEEE Conference on Secure and Trustworthy Machine Learning (SaTML 2026). Introduces the first systematic adversarial attack framework against model fingerprints in AI-generated images. Preprint available on arXiv.
• Sep 2025
  Released AuthPrint, a black-box ML approach for detecting silent model swapping by malicious image-generation API providers. This is the first work to formally address this threat model.
• Apr 2025
  Presented SoK: What Makes Private Learning Unfair? at SaTML 2025 (Copenhagen).

Publications

1. Smudged Fingerprints: A Systematic Evaluation of the Robustness of AI Image Fingerprints
   Yao K, Juarez M.
   arXiv preprint, 2025. To appear in 2026 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML).
2. AuthPrint: Fingerprinting Generative Models Against Malicious Model Providers
   Yao K, Juarez M.
   arXiv preprint, 2025.
3. SoK: What Makes Private Learning Unfair?
   Yao K, Juarez M.
   2025 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML).

Note: ^* denotes equal contribution (co-first author).

Teaching and Tutoring

• Teaching Assistant
  Privacy and Security with Machine Learning — University of Edinburgh (2023–2025)
• Teaching Assistant
  Mathematical Image Analysis — Johns Hopkins University (2019–2020)
• Research Tutoring
  Felipe Takaesu (JHU), Eliana Crentsil (JHU), Lucia Sablich (JHU), Shannon Flanary (JHU), Chunhan Fang (UoE).