darinor
All writing

Prompt injection & red teaming

How agents actually get compromised — indirect injection, cross-origin reads, and the social-engineering moves that beat an agent's own refusals without ever touching the model.

Articles in this collection

6 articles