The team behind WhatPay announced their AI-native multi-chain wallet last week. I spent Tuesday night digging through the available information. The GitHub repo? Empty. The audit reports? Non-existent. The team members? Anonymous. What I found instead is a textbook case of narrative-driven product publishing — a polished interface wrapped around a black box of unverified claims.
Let me be clear: I’m not dismissing the concept. AI-powered conversation-as-trading is a legitimate UX improvement. But after 21 years in this industry, I’ve learned that the gap between a demo and a secure product is where most user funds get lost. WhatPay hasn’t even acknowledged that gap.
Context: The State of AI Wallets
The wallet space is brutally competitive. MetaMask, Trust Wallet, OKX Wallet — each has millions of users, deep liquidity integrations, and years of trust. Into this arena steps WhatPay, offering a single differentiator: you can swap tokens by typing “Send 0.5 ETH to USDC on Arbitrum” in plain English. Behind the scenes, it uses MPC (Multi-Party Computation) for self-custody and claims to support 65 chains.
This is not a new layer-1 or a novel consensus mechanism. It’s an application-layer innovation — a wrapper around existing infrastructure. The technical risk isn’t in the blockchain; it’s in the AI backend that parses your intent and assembles the transaction. If that backend is compromised or hallucinates, you sign a transaction to a malicious contract.
Core: What the Code (Doesn’t) Tell Us
First, the MPC claim. MPC is a mature technology — Fireblocks and ZenGo have used it for years. But WhatPay hasn’t disclosed the threshold (2-of-3? 3-of-5?), who controls the shards, or how key recovery works. Without that, the phrase “self-custody” is just a marketing label. I’ve audited enough ERC-20 contracts in 2017 to know that undefined security parameters are the first sign of trouble.
Second, the AI transaction flow. The official description says “AI automatically completes intent recognition, data retrieval, and result generation.” But they don’t specify which LLM they use, how they structure on-chain data, or how they prevent hallucinations. During my 2020 DeFi arbitrage work, I learned that even a single incorrect token address can cost you the entire position. WhatPay expects users to blindly trust an AI that could be serving malicious data.
Third, the 65-chain support is almost certainly shallow. In the crypto world, “support” can mean anything from “read-only balance display” to “native swap and DApp connection.” Given the absence of any technical architecture document, I’m reasonably confident that the long-tail chains are only for querying, not for full interaction. This is a common pattern in early-stage multi-chain wallets — they integrate an RPC provider and call it support.
The Missing Economic Layer
WhatPay’s announcement contains zero information about tokenomics. No fee structure, no incentive plan, no governance token. That’s unusual for a product that expects to attract users. The only plausible explanation is that the project is in an extremely early stage, and the team is still deciding whether to launch a token. If they do, the value capture will depend on whether AI service fees or trading fee splits are tied to the token. Without that, the token is a pure speculation vehicle.
Contrarian: The Real Risk Is Not the AI, It’s the Trust Model
The conventional critique of AI wallets is that AI can hallucinate. That’s true, but it’s a solvable engineering problem. The deeper issue is that WhatPay’s architecture creates a single point of failure: the centralized AI backend. If the backend goes down, the wallet becomes unusable. If it’s hacked, the attacker can manipulate transaction parameters. The user’s only protection is a manual review of the transaction details — but the whole point of the AI is to reduce friction. Most users will click “confirm” without reading the tiny contract address.
This is the same dynamic that caused the Terra collapse: users trusted the algorithm without understanding the mechanics. WhatPay is asking users to trust an AI that they can’t audit, from a team they don’t know, running on infrastructure they can’t verify. That’s not innovation; it’s a new attack surface.
Takeaway: Watch the Incumbents, Not the Upstarts
WhatPay may capture a niche of early adopters, but its narrative advantage is temporary. MetaMask and OKX are already integrating AI features. When they do, they’ll have the liquidity, the user base, and the trust that WhatPay lacks. The real question is not whether WhatPay will succeed, but whether the AI wallet narrative will survive the inevitable security incident. When it does, the market will remember that the code doesn’t lie — and the absence of code is the loudest lie of all.
Arbitrage is just geometry disguised as finance. Trust is just risk management deferred. WhatPay has deferred the risk. I’ll wait until they publish the geometry.
