AI · 132 days ago

Cisco research demonstrates VLMs can be manipulated by imperceptible pixel perturbations embedding instructions in images (e.g., banners/previews), creating a covert channel that humans and simple filters miss.

Cisco research demonstrates VLMs can be manipulated by imperceptible pixel perturbations embedding instructions in images (e.g., banners/previews), creating a covert channel that humans and simple filters miss.

Timeline

• May 7 Reported by SecurityWeek

Sources

1 source covering this story

• SecurityWeek May 7

  Attackers Could Exploit AI Vision Models Using Imperceptible Image Changes

  Cisco’s AI security researchers have analyzed ways to target vision-language models (VLMs) using pixel-level perturbation.

Part of the PlainSec briefing for 2026-05-07

Editions

• Briefing for 2026-05-07