Poison-then-Hide: Finetuning-Activated Backdoor Attack on Pretrained Vision Encoders

# Poison-then-Hide: Finetuning-Activated Backdoor Attack on Pretrained Vision Encoders

https://gcdradar.vercel.app/incidents/poison-then-hide-finetuning-activated-backdoor-attack-on-pre · AIR-2026-1968 · live

- Lane: Research
- Severity: 1 (Research / proof of concept): Research paper demonstrating a novel backdoor attack on machine learning models.
- Fix status: Fix status unknown
- First reported: 2026-11-13
- AI role: Attack on AI
- Products: Vision Encoders

## Summary

Researchers introduce a stealthy finetuning-activated backdoor attack that implants a dormant backdoor into pretrained vision encoders. The attack bypasses limitations of existing methods under full-model finetuning.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-11-13: Poison-then-Hide: Finetuning-Activated Backdoor Attack on Pretrained Vision Encoders (https://arxiv.org/abs/2611.14970)

## Sources

- 2026-11-13 arXiv: Poison-then-Hide: Finetuning-Activated Backdoor Attack on Pretrained Vision Encoders https://arxiv.org/abs/2611.14970

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.