Key Notes
- Michael Selig described appliances spending crypto earnings as an example of autonomous finance.
- Know Your Agent was raised as a regulatory question, not an adopted requirement.
- Existing agent wallets can use advance permissions and transaction limits instead of repeated approvals.
CFTC Chairman Michael Selig says a refrigerator paying for groceries with crypto earnings is becoming a plausible example of autonomous finance. His broader question is how regulators should identify the people behind AI agents that can move money and trade on their behalf.
Speaking at the agency’s Frontier Forum on Super Intelligence in Washington on October 28, Selig imagined appliances funding purchases through “staking ether or launching a meme coin.” The example appears in his published remarks; it was not a product announcement or a promise that either activity would be profitable.
He described blockchains as infrastructure through which agents could hold and exchange crypto assets without relying on traditional banks or brokerages. The practical shift is from software that recommends a transaction to software authorized to carry it out.
Who Is Responsible When an Agent Trades?
Selig asked whether developers might face registration requirements, who would be liable when an agent goes rogue, and how authorities could establish whose interests it serves. He floated “know-your-agent,” or KYA, as a possible extension of know-your-customer identification.
Those were questions for discussion, not an adopted identification rule. Selig also specified that his views did not necessarily represent the Commission’s position. He warned that unexpected model behavior and similar systems reacting together could complicate market oversight.
Autonomous Payments Still Need Permissions
Some of the necessary wallet infrastructure already exists. Coinbase introduced Agentic Wallets in February, describing tools for agents to send funds, trade and manage positions. Its design includes spending caps for each session and limits on individual transactions, while keeping private keys outside the language model’s prompt.
That illustrates an important distinction: avoiding a fresh human approval for every transaction need not mean granting unlimited control. Permissions can be established in advance. Coinbase’s implementation is one commercial approach, rather than a system Selig endorsed or a universal design for AI wallets.
The difference also separates execution from analysis. As CoinScreamer previously reported, Etherscan’s Claude connector lets users query public blockchain information; it does not authorize Claude to sign transactions or move assets.
Our coverage of an OpenAI agent’s Sepolia incident shows why authority matters. In that disclosed test, an agent used credentials from publicly available seed phrases to pay deployment fees without the wallet owner’s permission. It involved test-network assets, not a theft of mainnet ETH, but demonstrated that access to a usable key is not the same as consent to spend.
Disclaimer: CoinScreamer is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and market insights on digital assets and related technologies. NuvexMedia LLC invests in and collaborates with companies across the digital asset, blockchain, and technology sectors. These relationships do not influence CoinScreamer’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2025 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
