When agents start holding wallets and acting for people, the question stops being ""can it do the task"" and becomes ""on whose authority, and within what limits."" This session looks at how to keep an agent bound to its principal's mandate, who it can pay, how much, under what conditions, while still letting it coordinate with agents acting for other humans. EigenCloud makes those limits enforceable instead of aspirational: TEE attestation proves the code that ran is the code that was meant to run, and sealed decision records make every action checkable rather than something you trust the operator about.

We'll look at what actually keeps an agent honest. The only rule that reliably stops an agent from being tricked into moving money is a strict list of approved recipients, and even that fails if it guards one way of sending funds while leaving another open. From there we'll ask what real agent sovereignty has to look like, so autonomous systems end up serving the many instead of the few.