safety security
The Blank Field Still Leaked Its Neighbors
Document models inferred missing identity fields from memorized relations when the input image no longer contained enough evidence.
Summary
Document models inferred missing identity fields from memorized relations when the input image no longer contained enough evidence.
The study tests identity-document key-information models under absent or minimal visual evidence and finds that correlated fields can still be inferred, exposing memorized relationships rather than observations. Its DocPrivacyBench measures that leakage, while the proposed DRUF unlearning method decouples high-risk field pairs and updates the target set dynamically. Across three multimodal models and six unlearning methods, DRUF improved leakage suppression by 4.8 percentage points over the strongest baseline while preserving extraction utility in the reported experiments.
Why it matters
Document models inferred missing identity fields from memorized relations when the input image no longer contained enough evidence.
Limits and context
No additional limitation was separately recorded.
Key claims
Document models inferred missing identity fields from memorized relations when the input image no longer contained enough evidence.
Evidence: source-2026-08-16-008
Sources
- arXiv preprint 2608.12911arXiv · primary research
Corrections
No corrections have been recorded for this story.