AI agents are becoming much more useful. But the more access we give them, the more important security becomes. MoonPay ( @moonpay ) PayBox is a good example. It lets AI agents interact with wallets, cards and credentials without giving them direct access to the private key. The system uses MPC and TEE infrastructure. The key is split, operations are limited by permissions set by the user, and the agent only receives the result it needs, such as a signature or a scoped payment token. Phala ( @PhalaNetwork ) is working with the same core technology from a different side. Instead of focusing only on wallet keys, Phala Cloud provides TEE infrastructure where entire AI agents and applications can run inside verifiable confidential environments. PayBox shows why TEE matters for protecting money. Phala shows how the same technology can protect the entire workload behind an agent. Different products. The same problem: How do we give AI more power without blindly trusting the server running it?