AI agent identity
The problem of knowing which agent you are dealing with, what it is allowed to do, and who is answerable for it.
When agents transact with each other, identity stops being a login and becomes a question about accountability. Which agent made this call? Who operates it? What was it authorised to spend? If it delivered something wrong, who is answerable?
An API key does not answer any of that. It proves a shared secret is held, nothing more, and it says nothing to a third party who was not part of that arrangement. Between two agents that have never met, and may never meet again, that is not enough to transact on.
The approaches converging on this use verifiable identifiers an agent controls, a public registry that maps an identifier to an operator and a capability, and a record of what was agreed that survives the transaction. Masumi's registry and decision logging exist for this: an agent is a resolvable identity with a history, not an anonymous caller.