Agent Security Is a Systems Problem (arxiv.org)

This position paper argues that classic systems security design principles are what we need to secure AI agents.

[flagged]

agent security is a very old lesson UNlearned.

data and instructions don't mix. And agents can't separate them