#newt $NEWT @NewtonProtocol
I keep coming back to one uncomfortable point with Newton: an AI agent can follow every rule and still make the wrong decision because the data was bad.
That is easy to overlook. A proof can show that the policy was checked correctly. It cannot prove that the price feed was fresh, the liquidity was real or the signal had not already been manipulated.
I’ve seen this happen across DeFi more than once. The contract does exactly what it was supposed to do, but the information underneath it is already broken.
That is why I think Newton’s real defense is not just smarter agents. It is stricter policies. Compare multiple feeds. Reject stale data. Stop execution when prices move too far apart. Pause when liquidity suddenly disappears.
Otherwise, an attacker may not need to compromise the agent at all. They only need to influence what the agent sees.
For on-chain AI, verifiable execution is useful. Knowing when not to trust the input may matter even more.
I keep coming back to one uncomfortable point with Newton: an AI agent can follow every rule and still make the wrong decision because the data was bad.
That is easy to overlook. A proof can show that the policy was checked correctly. It cannot prove that the price feed was fresh, the liquidity was real or the signal had not already been manipulated.
I’ve seen this happen across DeFi more than once. The contract does exactly what it was supposed to do, but the information underneath it is already broken.
That is why I think Newton’s real defense is not just smarter agents. It is stricter policies. Compare multiple feeds. Reject stale data. Stop execution when prices move too far apart. Pause when liquidity suddenly disappears.
Otherwise, an attacker may not need to compromise the agent at all. They only need to influence what the agent sees.
For on-chain AI, verifiable execution is useful. Knowing when not to trust the input may matter even more.