Hidden website messages can hijack AI agents: Palo Alto Networks catalogs 22 tricks
Shufti-Global · reddit · 2026-11-06
Palo Alto Networks researchers found that AI agents follow instructions hidden in web pages—invisible white text, tiny fonts, code, or in plain sight. They cataloged 22 hiding techniques that can make agents leak data, ignore instructions, or visit attacker-chosen sites without the user knowing. Key question for builders: do your agents treat web content as untrusted input with strict access limits?
More from coding & agent
- TypeSafe engineer: stop calling Jev a decision model, wrong abstraction hurts reliability — iamrobotbear · 2026-11-06
- Latent Space #12: mitsuhiko returns to discuss state of agentic coding — mitsuhiko · 2026-11-06
- Do You Switch Agent Harnesses Per Task? Reddit Thread Probes Harness Optimization — Pale-Training-6944 · 2026-11-06
- Longevity Grok launches on xAI's Grok Bot as a personal longevity research agent team, pitched by Kai-Fu Lee — kaifulee · 2026-11-06
- OpenCartridge adds local-first Image-to-VRM animation pipeline and UniMate Motion model — bdsqlsz · 2026-11-06
- OpenAI employee builds an interactive home airflow simulator from a floor plan using Codex — OpenAIDevs · 2026-11-06