Right now, the standard way to authenticate an AI agent is a long-lived API key. Those keys usually carry far more access than the agent needs, and nothing cryptographically ties the agent back to a person. If one leaks, that's a real problem.

Kite does it differently. Identity comes in three layers: user, agent, session. You hold the root. Your agent gets its own address, derived from your key. Permissions are set per task, not per agent.
The part I like is the delegation. Authority gets passed down with cryptographic proofs, so nobody has to just trust a setting. Even if an agent is fully compromised it can't go past the limits you set. A service can also check that an agent is backed by a real human without ever learning who that human is.

It's still early and what matters now is whether builders actually use it. But the problem it targets is real.

#KiteBBW @KITE AI $KITE