> the issue of the key leaking is still a problem with your proposed protocol Fair enough. I intended that it's not a long term key, but something more approp

> the issue of the key leaking is still a problem with your proposed protocol Fair enough. I intended that it's not a long term key, but something more appropriate like a session key. Or better, more like a "query key" that limits what can be extracted from a session to whatever has been approved to be extracted. (See "zero-knowledge database".[1]) > What you're looking for is a protocol where N parties have to agree in order to decrypt any given message, and agreement to decrypt a particular message doesn't allow them to decrypt any other messages That's what I meant, yes. Sorry for not making that clear. > Same problem with giving the key only to an AI: The AI needs the key, and it will be difficult to ensure that the AI system isn't hacked Ah... The other thing I didn't make clear is that the AI runs inside homomorphic encryption[2], or other protective bubble against access (one can imagine a quantum state with this property). This is why I said we don't have the technology to do it yet. Not because of the AI, but because we don't have sufficiently powerful methods to run an AI (or any large program) inside a bubble that prevents them from being inspected. But we know it's possible in principle. [1] https://www.cs.jhu.edu/~susan/600.641/scribes/lecture15.pdf https://www.cs.jhu.edu/~susan/600.641/scribes/lecture15.pdf [2] https://en.wikipedia.org/wiki/Homomorphic_encryption https://en.wikipedia.org/wiki/Homomorphic_encryption