AI trading bots can buy and sell faster than any human. That speed is exciting, but it also raises a tough question for fund managers: how do you keep a robot trader from going off the rails without revealing your secret strategy? A September 7, 2026 post from the Midnight team proposed an answer called Policy Vaults.
The Dilemma
Fund managers face two bad options. They can let a bot trade unmonitored and risk a large loss. Or they can make its orders and risk settings public, which lets competitors copy or trade ahead of it, a practice known as front running.
The post points to real world examples of what happens when risk limits fail, including a mistyped order at Citi that triggered a $444 billion basket error, and the $5.5 billion Archegos collapse that followed risk limits not being enforced.
How a Policy Vault Works
Money sits in a smart contract, a program on the blockchain that runs automatically, with risk limits written directly into its code.
The bot cannot touch the funds directly. It must send each trade to the vault.
The vault only carries out the trade if it fits the rules, such as a maximum $100,000 position or a $500,000 daily loss cap.
Using zero knowledge proofs, the vault can prove that every trade followed the rules without revealing the trades themselves. Think of it as a referee who confirms the game was fair without showing anyone your playbook.
Proof of Control
The post also introduced a "Proof of Control" framework with six properties: privacy, verifiability, security, identity, portability and provenance (a record of where something came from). It argues that Midnight Policy Vaults could combine all six, while today's custodians, public blockchains and credential standards each cover only some.
Why It Matters
Institutions want both oversight and confidentiality. Tools like Policy Vaults show how Midnight aims to deliver both, which could open the door to more professional capital using privacy focused blockchains.