Researchers Detail Finetuning-Activated Backdoor Attack on Pretrained Vision Encoders
A new paper introduces 'Poison-then-Hide', a backdoor attack on pretrained vision encoders that stays dormant until the model is finetuned, letting attackers control model behavior via triggers.
Read the full story at arXiv cs.LG ↗
Timeline · 1 report
- 2026-11-13 05:00 · arXiv cs.LG
Poison-then-Hide: Finetuning-Activated Backdoor Attack on Pretrained Vision Encoders