AI Agents Are Democratizing Finance but Also Redefining Risk

This report examines the security risks of AI agents that autonomously move crypto capital, explaining how reliance on unverified external inputs and possession of private keys can enable manipulation or theft (e.g., via prompt injection or compromised integrations). It recommends splitting authority and enforcing policies outside the agent (for example, using MPC and independent control layers) so agents can execute strategies without having unilateral control of funds.