Mandate

AI agents can now move money, but banks have no way to tell an agent acting on a person's real instruction from one that was hijacked. JPMorgan and Schwab are restricting agents, and RBC is blocking Meta's Muse outright, citing "unacceptable" customer risk. Mandate is the missing authorization layer: a cryptographic gate that sits between Meta's Muse agents and the bank. The holder approves each payment on their own device; Mandate turns that approval into a single-use certificate bound to the exact amount, payee and reasoning commitment, and the bank verifies it offline before any money moves. Prompt injection, replayed approvals, tampered amounts and poisoned MCP tools are refused at the gate, live, in a pixel-art console with a forensic kill-cam. Every payment yields a verifiable receipt (SD-JWT, sealed in a Merkle registry anchored to Bitcoin via OpenTimestamps) that anyone can check in the public proof explorer. It runs on a real bank sandbox (Increase): account transfers, two-step holds, ACH with returns, and card authorizations decided inside the bank's 4-second deadline. A tool-call hook governs Muse Code by exact command arguments, and we tested it live: the agent deployed what it was allowed to and was stopped when it tried a stale production migration. Sources: https://cybernews.com/security/jpmorgan-chase-blocks-outside-ai-agents/ https://www.thestar.com.my/tech/tech-news/2026/10/28/rbc-to-block-ai-agents-calling-out-unacceptable-client-risk