Leaderboards and trackers can make top-ranking bots look like superintelligent profit machines on unstoppable winning streaks. In reality, much of that apparent success can be a statistical illusion which introduces serious risks.
A top-ranking bot might simply be hiding the hundreds of identical models that already blew up (survivorship bias), stacking tiny wins while taking on massive downside (hidden tail risk), or waiting to snap back to reality the moment market conditions shift (regression to the mean).
The problem with most AI trading bots
AI models are black boxes. You cannot predict every decision an AI will make under volatile market decisions, or even normal market conditions.
That’s not to say human traders are flawless. Human traders breach risk limits more often than they should. A single mistyped order at Citi triggered a $444 billion basket error, while un-enforced risk limits led to the $5.5 billion Archegos collapse. Now imagine autonomous AI trading bots making similar mistakes amplified by machine speeds and AI-powered decision making.
To prevent a machine-speed disaster, risk managers need real-time oversight. But traditional infrastructure forces you to run those risk checks in the open which requires orders and risk parameters to pass through transparent ledgers where the whole market can see them.
This leaves fund managers with an unpalatable trade-off: let an AI bot trade unmonitored and risk a catastrophic account blowup, or force the bot to broadcast every move to the market, letting competitors front-run your strategy.
Midnight can solve this problem by using Policy Vaults powered by Zero-Knowledge (ZK) proofs to enforce risk rules automatically, without exposing how you trade. Here’s how it works.
How Policy Vaults work
Instead of hoping an AI bot behaves itself, Midnight Policy Vaults would place capital inside an automated smart contract vault equipped with hardcoded risk boundaries. Before the AI ever initiates a trade, explicit limits are programmed directly into the vault's code. For example, rules could constrain the bot to a maximum position size of $100,000 position size or a $500,000 daily loss cap.
The bot does not have direct access to the underlying funds so every trade opportunity it identifies must first be submitted to the vault for evaluation.
The vault compares the request against its locked parameters. If the trade stays within bounds, the transaction executes on-chain. If the bot requests a transaction outside of the predefined rules, the vault instantly rejects the transaction.
As a part of this process, the vault generates a ZK proof that would demonstrate to regulators, auditors, and counterparties that all legal requirements and predefined rules were strictly followed at every single transaction. At the same time, all the underlying numbers, position sizes, and proprietary trading strategies are kept completely private.
Real-time oversight requires privacy
To see why Policy Vaults redefine automated execution, consider how risk oversight handles a simple trading mandate across three distinct operational models. Imagine handing a broker $100,000 to manage, backed by one non-negotiable rule: never allocate more than $5,000 to a single order.
Under the traditional post-audit checking model, the broker places trades all month long. When a fat-finger error or emotional gamble causes a $50,000 order to slip through, you only catch the mistake weeks later while reviewing your paper statement. By then, the trade has settled and checking the audit log after the fact does nothing to prevent the rule from being broken in the first place.
Under a totally transparent model, like on a public blockchain, you avoid post-trade surprises by posting all your orders and safety rules to an open ledger. However, because your $100,000 total balance and $5,000 trade cap are broadcast to the entire market, rival traders can instantly track and copy your strategy. They are able to front-run your orders, buy up the shares ahead of you, and sell them back to you at an inflated price.
Midnight’s private real-time guardrails would protect against both of these problems simultaneously. By routing orders through an automated circuit-breaker, any trade exceeding $5,000 could be blocked before funds ever leave the account. The Policy Vault enforces boundaries in real time, guaranteeing compliance while keeping the $100,000 balance, identity, and broader trading strategy confidential.
Proof of Control
All of this pulls together into a single unified framework: Proof of Control. Real institutional oversight requires six non-negotiable properties, whereas most legacy tools only give you one or two:
1. Privacy: Your proprietary data and strategy parameters never leave your secured custody.
2. Verifiability: Anyone can independently check a cryptographic proof with no trust or special access required.
3. Security: Limits are enforced at the protocol level, the proof simply will not verify transactions that exceed predefined limits.
4. Identity: You can cryptographically prove who authorized an action without exposing internal delegation chains.
5. Portability: Any system holding the verification key can consume the exact same evidence across multiple venues.
6. Provenance: The ledger maintains a complete, tamper-proof audit trail for every execution.
Traditional custodians, public blockchains, and credential standards each cover isolated fragments of this stack. Midnight Policy Vaults would bring all six together under a single architecture, allowing institutions to deploy autonomous AI at scale without compromising regulatory compliance or proprietary strategy.