FARE: Forensic Acceptance Region Estimation for Catching Bait-and-Switch Image Generators
FARE flags silent swaps of certified AI image generators by testing images against enrolled forensic artifacts.
Researchers propose FARE, a deployment-time auditor that enrolls a certified image generator from sampled images and later checks whether a new image is consistent with that generator. It amplifies generator-specific forensic artifacts by training on hard samples that tighten the acceptance region. Across substitutions, including similar versions and variants, FARE outperformed baselines at strict operating points and remained effective under the exact-model and decision-only attacks evaluated in the paper.
- FARE enrolls a certified generator using only sampled images
- Hard-sample training tightens the forensic acceptance region
- It detects swaps among similar model versions and variants
- It beat baselines at strict thresholds under the tested attacks
From the original article
1 min read at arxiv.orgModern AI image generators are increasingly deployed as opaque APIs, where customers can query the deployed service, but cannot inspect model weights or architecture. This creates a practical challenge: a provider may pass governance certification with one generator and later silently switch to a cheaper and lower-quality one for deployment, compromising public trust or even safety in high-stakes domains.
By Kai Yao · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.