At 2 a.m. the phone vibrated. The recorded voice from my brother carried familiar breaths and a half-laugh: “My ticket money was scammed away outside the Kansas City stadium. I need to reroute $400 right now—the Argentina knockout match is about to start.” The voice was exactly like his. Every pause sounded real. In the end, this plea for help was confirmed to be an AI-cloned voice. The scammers recreated a person’s timbre in just a few seconds using social audio. That time I didn’t send the money, but I learned this: trust can’t be built on “it sounds like it.” What you receive is a voice, not authorization.

In 2025, Americans lost more than $893 million to AI-related scams. Among them, emergency family voice scams are the fastest-growing category. Suyay’s experience ranks No. 4 on today’s trending list; a 1,200% growth rate has even led the FBI to classify this kind of attack as high risk. But when we manage a treasury on-chain, we still use a similar decision logic: once a transaction comes from a familiar contract address, we assume the rules have been followed. The truth is, on-chain only records what happened—it never says why it was allowed.

The problem is stuck in the order of the validation logic. On-chain transactions only do what, not why. Multi-signature wallets may have added human review, but it’s basically asking a few big shots to listen to the “audio” again—if it sounds right, it gets approved. Without machine-readable policy and proof-of-compliance, internal violations or vulnerabilities are only discovered after the fact, at an utterly outrageous cost. What we need isn’t a more sensitive AI detector, but a clear permission path from the very beginning.

Before a transaction enters execution, the Newton Protocol inserts a layer of Rego—plainly put, it’s a programmable policy checkpoint. If you want to move funds from the vault, the system first runs the rules: Is the amount over the limit? Whitelist address? Time window? Only if everything passes will it generate an approval attestation, attaching the policy and the operator’s signatures for that operation, and then the transaction is approved. This setup doesn’t require you to go check the GitBook again—every $NEWT transfer comes with a numeric credential for “why it’s allowed.” The rules are truly encoded into the instructions, not left on paper.

Now you can go to the Newton Mainnet Beta and freely look up an attestation record. The key is just one thing: whether it’s linked to a specific policy ID and the operator’s signature. If the entry you find is missing any one of those, then the so-called vault policy is still just a trust gap at the “voice-clone” level. @NewtonProtocol