AI agents are starting to make decisions with real consequences. But how do you prove an agent stayed within the authority it was given?
Logs alone aren’t enough. If the operator controls the records, how can anyone independently verify what actually happened?
That’s the problem @Mysten_Labs is addressing with the Verifiable Agent Arbiter (VAA), developed in collaboration with Google Cloud.
The idea is simple: keep detailed agent records, including prompts, outputs, tool calls and policy decisions, private in Google Cloud Storage, while storing cryptographic proofs on @WalrusProtocol and coordinating them on @SuiNetwork .
This means the underlying records can remain private while their integrity can be independently verified. If a record is altered, its proof will no longer match.
VAA also connects to Sui Agent Payments, bringing an agent’s spending decisions and authorization into the evidence trail.
As AI agents gain more autonomy, accountability cannot depend on what an operator says happened.
We need evidence that can be independently verified, not just logs we are expected to trust.
🤲🏾 Bitget Hackathon S2 voting is live.
Go show S151 some love. I built Reckon, an AI tool that stress-tests trading decisions before you take a trade.
Check it out: https://t.co/fKn1OPKrpl
To vote, drop S151 in the comments under the official tweet I quoted.
Let’s see if we can bring this one home.
Find out who is launching on @blastdotfun.
Me and @Cussysss put the teams in the hot seat.
no slides, no corporate talk, just vibes..
https://t.co/3B7roSlwyf