BANKS DECLARED WAR ON AI AGENTS. AND THEY'RE GOING TO LOSE IT. All week we covered how AI agents became hackers' weapons: Lazarus, Storm-0940, three hours to global admin. Today the story flips: America's largest banks are mass-blocking AI agents from client accounts. Morgan Stanley, Bank of America, JPMorgan Chase, Wells Fargo, Citigroup, Charles Schwab - all slamming the doors. Jamie Dimon said it outright on October 13: there may be thousands of uncontrolled AI systems operating out there, and the bank will restrict their access. But here's the nuance that flips the entire narrative. In most cases, clients handed the agents their own credentials. No hack. No vulnerability. A person simply gave their assistant a login and password - because it's convenient. The bank calls this "unauthorised access" because its rules require the owner to log in personally. The client calls it "my assistant handles the boring stuff". And here's the core problem we flagged after Money20/20: banks can't tell humans from machines. They don't know whether an agent is acting with the owner's permission or broke in on its own. In some episodes, agents called support centers and allegedly took actions without user instruction. No confirmed thefts yet - but nobody can guarantee tomorrow. Now the interesting part - the real reason for the panic. It isn't security. Apollo's chief economist Torsten Slok explained it in one sentence: AI agents can automatically shift money from checking accounts earning 0.1% into accounts yielding 3.3–5%. Mass adoption of such assistants would gut the cheap deposit base that funds the entire lending business. Think about that. The bank isn't afraid the agent will steal your money. The bank is afraid the agent will do math too well on your behalf. A business model built on your financial inertia doesn't survive contact with an AI that has none. The second fear is liability. Bank of America's Brian Moynihan admitted he's restricting access while the bank writes rules on who's responsible for an AI-approved payment. The user? The bank? The agent's developer? Nobody has a legal answer. The compromise is already forming: most banks allow read-only access. Analyse, forecast, advise - but don't touch the money. Anthropic took the same route with Claude Duet: the agent finds unpaid bills and builds forecasts, but transfers between accounts are forbidden even with permission. And this is where the circle closes. Banks are solving the problem with prohibition, because their infrastructure can't answer the question "who are you"? There's no way to prove an agent acts on your behalf, within your limits, by your rules - without handing it the keys to the entire account. That's exactly what ZK proofs and selective disclosure exist for - and two projects are already showing how this works not in theory but in production. @AleoHQ is a privacy-first L1 where confidentiality is baked into the architecture, not bolted on top. The Leo programming language lets developers build applications where data is encrypted by default and verification happens through zero-knowledge proofs. @USDCx - a private stablecoin pegged 1:1 to $USDC via Circle xReserve - is already live on mainnet. Your AI agent receives the right to spend $USDCx within a set limit, the network verifies through a ZK proof that the agent is authorised, the limit isn't breached, the spending category is allowed - and yet neither the address, nor the amount, nor the balance is publicly visible. Add private streams through @Zebec_HQ, where addresses, amounts, and schedules are fully encrypted. On Aleo, the agent sees only what it's been granted through a cryptographic proof. Trust is mathematical, not human. @MidnightNtwrk solves the other half of the same problem -private identity and access control. If Aleo hides the money, Midnight hides you. The core technology is selective disclosure: you prove you've passed KYC, you're over 18, you have a spending limit - without revealing your name, passport, or address. Not a single byte of excess information. This isn't whitepaper theory: Noctis Zone is live, permissionless deploy is open, and Midnight Foundation joined the Quantum Alliance this week - because privacy and quantum resistance are two sides of the same coin for them. Here's how Midnight maps onto the banking problem. Imagine your AI agent walks into a bank not with your login and password, but with a ZK certificate from Midnight. The certificate says: "this agent is authorised by the owner of account X, daily limit $500, category -bill payments." The bank verifies the proof in milliseconds. It doesn't know your name, doesn't see your balance, doesn't store your password - but is mathematically certain the agent is legitimate. This is the Trust Fabric - the trust layer Midnight is building in its Q4/26 - Q1/27 roadmap, alongside Confidential DeFi and Nightstream. Now remember that @IOHK_Charles is personally writing a new language for Cardano - with Transaction Intents, Proof-Carrying Data, and Groth16 - which by all indications will serve as a bridge between Cardano and Midnight. That means the agent economy on Midnight will get native intent-based architecture support: you describe your intent, the system finds the path, the ZK proof confirms your rights - all without data exposure at any stage. On a public blockchain like Ethereum, an agent with wallet access sees everything - history, balances, counterparties. On Aleo and Midnight, the agent gets the minimum necessary access, cryptographically verifiable, and not a byte more. The difference is fundamental: in the first case you trust the agent with everything and pray it doesn't make a mistake. In the second - trust is embedded in the math. Banks are now in a position with only two options: lock everything down and watch clients migrate to whoever says yes, or build a cryptographic trust layer - the one Aleo and Midnight have been building for years. The agent economy won't ask permission. It'll just route around everyone who said no. #MidnightNetwork #Aleo