Ohh that's a good idea.
If I'm understanding this correctly you proactively filter comms through an AI. If it detects you said something that could be interpreted a certain way the communication isn't sent and the user is told to be more careful in their phrasing?
That's pretty brilliant to be honest. Except maybe the discovery will then be on what the AI filter was instructed to prevent as if that's an admission of guilt.