If we don't already have it now, it's only a matter of time before we have a combination of deep-fake AIs with conversational AIs which can break a voice call m

If we don't already have it now, it's only a matter of time before we have a combination of deep-fake AIs with conversational AIs which can break a voice call mid-stream and begin influencing the conversation in a particular direction. Think of it like a man-in-the-middle attack. Our intelligence agencies would be absolutely stupid not to work on this. Any voice communication channel you can intercept, you can influence, and you can use the trust established with vocal familiarity to enable you assume the influence level of one or both of the speakers. The key would be training the AI to know when to alter the conversation and when to let it pass through.