Defending AI Agents
עברית Join waitlist
01Blog

Notes from the lab

Technical posts on securing AI agents: attacks we reproduce, defences we build, and what they score when measured.

  1. Payload splitting: when the detector and the model read different pagesAn attacker cuts an injection in the middle of a token and puts one unrelated line between the halves. Every pattern goes blind; the model reads straight through. The fix comes from network intrusion detection.By Shimon R. Malka · 10 min read

Atom feed