The "Ghost Agent" Problem In Newton’s Model Registry
I’ve been looking at Newton Protocol’s Model Registry and I think there’s one problem we should address early. I’m calling it the "Ghost Agent" problem.
The Model Registry is where developers list AI agents. To list an agent you pay a registration fee in NEWT. Operators also stake NEWT to execute tasks. The idea is simple. Good agents earn fees. Bad ones get penalized. Over time the market weeds out poor services.
But here’s the gap I see. What if I pay the fee and never run the agent? I don’t stake an operator. I don’t execute any tasks. I just leave it listed in the registry.
Why would someone do that. To occupy a name. To create noise. To make it harder for real agents to be found. I’ll call this Agent Squatting.
Right now I don’t see any public rule that says delist an agent if it isn’t used. So it can sit in the registry forever with zero executions. Slashing only happens if there is an operator and a failed task. With no activity, there is nothing to penalize.
My proposal is Proof-of-Usage.
If an agent has zero executions for 90 days, auto delist it from the registry.
90 days feels fair. It gives developers time to find users, but it stops people from squatting forever.
This could work by tracking last_execution_timestamp on-chain. After 90 days of inactivity, remove the listing. No fee refund, so spamming becomes expensive. You can re-list anytime by paying the fee again.
This does not close the registry. It just keeps it clean. Users see agents that are actually being used. Operators get a better signal. And squatting becomes costly.
Newton wants to be the coordination layer for on-chain automation. For that, the registry should reflect agents that are actually working, not just agents that paid a fee once.
This is just my take based on how the registry is designed today. But I think Proof-of-Usage is a small rule that could prevent a big problem as the marketplace grows.@NewtonProtocol #newt $NEWT
I’ve been looking at Newton Protocol’s Model Registry and I think there’s one problem we should address early. I’m calling it the "Ghost Agent" problem.
The Model Registry is where developers list AI agents. To list an agent you pay a registration fee in NEWT. Operators also stake NEWT to execute tasks. The idea is simple. Good agents earn fees. Bad ones get penalized. Over time the market weeds out poor services.
But here’s the gap I see. What if I pay the fee and never run the agent? I don’t stake an operator. I don’t execute any tasks. I just leave it listed in the registry.
Why would someone do that. To occupy a name. To create noise. To make it harder for real agents to be found. I’ll call this Agent Squatting.
Right now I don’t see any public rule that says delist an agent if it isn’t used. So it can sit in the registry forever with zero executions. Slashing only happens if there is an operator and a failed task. With no activity, there is nothing to penalize.
My proposal is Proof-of-Usage.
If an agent has zero executions for 90 days, auto delist it from the registry.
90 days feels fair. It gives developers time to find users, but it stops people from squatting forever.
This could work by tracking last_execution_timestamp on-chain. After 90 days of inactivity, remove the listing. No fee refund, so spamming becomes expensive. You can re-list anytime by paying the fee again.
This does not close the registry. It just keeps it clean. Users see agents that are actually being used. Operators get a better signal. And squatting becomes costly.
Newton wants to be the coordination layer for on-chain automation. For that, the registry should reflect agents that are actually working, not just agents that paid a fee once.
This is just my take based on how the registry is designed today. But I think Proof-of-Usage is a small rule that could prevent a big problem as the marketplace grows.@NewtonProtocol #newt $NEWT
