Autonomous AI agents are increasingly trading and moving funds without step-by-step human approval, sharpening a legal debate over who is responsible when something goes wrong. Edwin Mata, lawyer and CEO and co-founder of tokenization platform Brickken, said liability should attach to the person or institution that authorized the software and defined its powers, not to the AI itself, because current law does not treat AI as a legal person. He argued that losses from trades executed within an approved mandate would ordinarily fall on the principal, while developers, platforms or financial institutions could face claims if flawed design, weak controls or corrupted inputs pushed an agent beyond its intended role. The debate is becoming more urgent as AI agents gain direct access to wallets, trading systems and payment rails. Mata pointed to ERC-8226, a draft Ethereum standard also known as RAMS, as one approach to making delegated authority verifiable onchain through time limits, spending caps, revocation controls and usage records. Existing U.S. rules already keep responsibility with regulated firms: SEC Rule 15c3-5 requires broker-dealers to control automated market-access systems, and consumer payment rules under Regulation E and CFPB guidance require authenticated authorization and a way to stop future transfers. Regulators and lawyers still lack a single answer for all autonomous-agent losses, however, meaning courts may continue to assess liability case by case based on control, mandate and system failure.