Sleeping Secrets: How Fine-Tuning Reawakens Privacy Risks in Language Models

Name: Sleeping Secrets: How Fine-Tuning Reawakens Privacy Risks in Language Models
Uploaded: 2026-10-05T05:15:59.351909+00:00
Duration: 468 s
Description: Researchers developed ReGap, a data-free attack to recover previously learned private associations from fine-tuned language models without access to original private data. By generating model-created candidates and selecting them based on likelihood under the frozen model, ReGap achieved significant recovery gains of 6–21 percentage points over the base model.

Researchers developed ReGap, a data-free attack to recover previously learned private associations from fine-tuned language models without access to original private data. By generating model-created candidates and selecting them based on likelihood under the frozen model, ReGap achieved significant recovery gains of 6–21 percentage points over the base model.

Open the full episode