The most interesting thing about Newton Protocol is not that it wants AI agents to automate crypto. That part is expected now.
The real claim is bigger: AI agents should be able to act for users without being fully trusted.
That matters because the moment an agent touches a wallet, the risk changes. A normal DeFi transaction happens once. You sign, approve, swap, bridge, or deposit. With an agent, you are not just approving one action. You are allowing future behavior.
That is delegation, and crypto has never handled delegation very well.
Newton’s idea is to make that delegation more limited. Instead of giving an agent full wallet access, users can define what it is allowed to do. It can spend only a certain amount. It can interact only with approved contracts. It can follow a strategy, but not move outside its boundaries.
That is useful. If an agent is managing yield, it should not touch principal. If it is trading, it should not send funds anywhere it wants. If it makes a mistake, the damage should be contained.
But this is where the promise gets complicated.
A permission system is only as good as the rules inside it. A rule can stop an agent from doing something clearly forbidden. It cannot always stop the agent from doing something technically allowed but still stupid.
A spending cap can prevent a full wallet drain, but it cannot guarantee a smart trade. An allowlist can block unknown contracts, but it cannot prove approved contracts are always safe. A policy can prove the agent followed instructions, but not that the instructions were wise.
So Newton does not remove trust completely. It moves trust into the limits, the defaults, the data, and the people who design the guardrails.
That may still be a major improvement. Crypto does not need AI agents with unlimited confidence. It needs systems that can tell them no.
Newton’s real value may be in taking that boring part seriously.
But the hard question remains: if users are not writing the rules themselves, who decides where the walls go?
