When your image is the attack vector, every picture becomes a potential payload. Technical walkthrough of visual prompt injection against multimodal LLMs including GPT-4V and Gemini Vision.

Invisible Commands: Visual Prompt Injection Against Multimodal LLMs

Secure By Dezign is the AI Security Training Ground for security architects, CISOs, and pentesters who need to stay ahead of adversarial AI. Every episode delivers a deep-dive into a real AI attack technique — how it works, how it's been exploited in the wild, and exactly how to defend against it. Hosted by Pax.

Share Invisible Commands: Visual Prompt Injection Against Multimodal LLMs

Sign up to save your podcasts

Invisible Commands: Visual Prompt Injection Against Multimodal LLMs

Invisible Commands: Visual Prompt Injection Against Multimodal LLMs