#newt $NEWT Most people talking about AI risk seem obsessed with one thing.
What if the model makes the wrong decision?
I used to think that was the big problem too.
Then I spent some time reading Newton Protocol, and I got stuck on a different question.
What if the AI makes the right decision... but it was never supposed to make that decision in the first place?
That honestly feels like a much more realistic failure.
An AI can generate the correct transaction. It can find the best route. It can optimize everything perfectly.
None of that answers one simple question.
Who actually gave it permission?
I think people mix up intelligence with authority all the time.
They're not the same thing.
A model can be incredibly smart and still have no business touching a particular wallet, moving a specific asset, or approving a certain transaction.
That's what stood out to me while reading Newton Protocol.
The protocol doesn't assume every technically valid transaction should be executed. It checks whether the action matches the policies defined beforehand. Spending limits. Delegated permissions. Identity requirements. Whatever rules the application decides to enforce.
The authorization comes before execution.
That was the point where my thinking changed.
Maybe the real challenge isn't building an AI that never makes mistakes.
Good luck with that.
Maybe it's making sure an AI can't act outside the authority it was given—even when it's technically capable of doing so.
One thing kept coming back to me while reading the documentation.
Intelligence answers "Can I do this?"
Authorization answers "Am I allowed to do this?"
Those aren't the same question.
Maybe I'm wrong, but if AI is going to handle real money, I'd want
permission checked before intelligence every single time.
#Newt @NewtonProtocol $NEWT
What if the model makes the wrong decision?
I used to think that was the big problem too.
Then I spent some time reading Newton Protocol, and I got stuck on a different question.
What if the AI makes the right decision... but it was never supposed to make that decision in the first place?
That honestly feels like a much more realistic failure.
An AI can generate the correct transaction. It can find the best route. It can optimize everything perfectly.
None of that answers one simple question.
Who actually gave it permission?
I think people mix up intelligence with authority all the time.
They're not the same thing.
A model can be incredibly smart and still have no business touching a particular wallet, moving a specific asset, or approving a certain transaction.
That's what stood out to me while reading Newton Protocol.
The protocol doesn't assume every technically valid transaction should be executed. It checks whether the action matches the policies defined beforehand. Spending limits. Delegated permissions. Identity requirements. Whatever rules the application decides to enforce.
The authorization comes before execution.
That was the point where my thinking changed.
Maybe the real challenge isn't building an AI that never makes mistakes.
Good luck with that.
Maybe it's making sure an AI can't act outside the authority it was given—even when it's technically capable of doing so.
One thing kept coming back to me while reading the documentation.
Intelligence answers "Can I do this?"
Authorization answers "Am I allowed to do this?"
Those aren't the same question.
Maybe I'm wrong, but if AI is going to handle real money, I'd want
permission checked before intelligence every single time.
#Newt @NewtonProtocol $NEWT