Recently, I was studying active wallets on a certain blockchain. At first, it was quite pleasing—positions were disciplined, entries and exits were orderly, and there was never any reckless chasing of hot trends. It was a textbook example of risk management.

But suddenly, a detail made me cautious. An operation itself was not excessive, but it carried a strange aura—like turning on the turn signal while the car subtly drifts to the side without realizing it. At that moment, I had a crazy thought: if this wallet is controlled by an AI proxy, who is the real decision-maker?

This question may seem absurd, but it hits the core contradiction in the crypto world. In the on-chain universe, the private key is like the steering wheel. It is the only key that can sign transactions. Every transfer and contract call requires a private key signature—similar to signing a check to confirm. Many users, for convenience, directly give their private keys to AI proxies for management. Once the proxy is hacked, malfunctions, or receives malicious instructions, the wallet can be looted in an instant. Imagine that helpless feeling.

Recently, I came across the idea proposed by KITE (KITE). Honestly, it was a revelation. Their solution is simple yet deadly: using two keys to divide authority.

The first is called the Human Key, which holds the root authority and control. It must be kept in an absolutely secure place and is not used regularly. The second is called the Agent Session Key, designed for temporary use by the AI proxy. The session key is like a temporary access card—check-in at work, and after finishing, swipe the card to leave. The permissions on the card are limited, and it automatically expires after a set time.

Some might ask: doesn’t this mean signing twice? Actually, no. The role of the Human Key is to set rules—creating, updating, or revoking session keys at any time—and to impose various fine-grained restrictions on the session keys. For example, daily transaction limits, asset types allowed, permitted counterparties, and so on. Permissions are tightly controlled from the source, so even if the AI proxy behaves abnormally, the scope of damage is firmly confined within the preset boundaries.

The brilliance of this logic is that it allows the AI proxy to work efficiently while ensuring users retain absolute control over their assets. Private keys are not a black-and-white choice; through meticulous division of authority, technology and trust can coexist.