>You can define a communication protocol between agents that fails when the communicating agent has been prompt injected
Good luck with that.
Yeah, how exactly would that work?
Yeah, how exactly would that work?