Much needed updates.
1. Hemlock was presented at CCS in Oct. We are working on releasing a usable tool in coming weeks.
2. Our next paper is peer-reviewed & accepted by USENIX Security 2026. It is a Model Provenance tool that computes "feature fingerprints" from genAI image models.
Model provenance tool distinguishes models trained from scratch from those that are "fine tuned" from existing models, and identifies any such "parent models." We will releasing this in coming months & distribute to folks in CA and IL legislative groups. Paper to be presented in Baltimore, Aug2026.
In terms of research, there is much more work in the pipelines, including a big study of AI music on consumer platforms that should raise quite a few eyebrows. That's coming early in 2026.
The paper:
Identifying Provenance of Generative Text-to-Image Models.
Anna Yoo-Jeong Ha, Wenxin Ding, Stanley Wu, Shawn Shan, Haitao Zheng, Ben Y. Zhao
To Appear: Proceedings of USENIX Security Symposium,
Baltimore MD, August 2026.
We're working on camera-ready version. Will post here when done.
Dec 9, 2025 13:51Have to admit, between research, work on legislative bills, work on legal cases, we sometimes forget about updating social media. Thx to folks here for reminding us :)
