In the future, AI will be able to operate as your agent. To do so will require near-omniscient access to your data: passwords, private keys, etc.
Can this be done privately?
TEEs and FHE offer privacy-preserving ways to implement an AI agent, but they are unproven. Nobody has been able to implement a full application backend on these technologies yet, much less a full-featured AI agent.
There’s a low-tech alternative: user-run backends.
Comm calls these user-run backends “keyservers”, and our thesis is that they’re going to get popular in crypto circles first.
Running nodes and dealing with private key cryptography is natural to crypto people.