Autonomous financial agents: Michael Selig opens debate on accountability

Software able to make decisions and act

Michael Selig, chairman of the Commodity Futures Trading Commission, devoted his October 28, 2026, remarks to the development of artificial intelligence agents in finance. His speech opened the first Frontier Forum held in Washington on these technologies. It sets out perspectives and regulatory questions with an explicit qualification: the views expressed are his own and do not necessarily represent those of the Commission. That distinction prevents each example from being read as an official decision already binding on businesses.

The CFTC chairman describes a shift from automation toward greater autonomy. Tools can analyze data, identify market opportunities and execute transactions for a user. Blockchains also enable them to hold and transfer digital assets without systematically relying on traditional intermediaries. As an illustration, Selig even discusses a household appliance funding purchases through ether staking. That scenario is not presented as an approved service or a recommended product.

Identifying the agent and the person it represents

The remarks raise several questions: should a developer register when an agent participates in a market? Who is accountable for software that acts unexpectedly? How can the person on whose behalf it trades be identified? Selig discusses a possible extension of know-your-customer checks to knowing the agent, sometimes referred to as KYA. The speech frames these points as issues to examine; it does not, by itself, establish a new registration procedure.

The approach he advocates should consider the functions performed, effective control and the economic reality of transactions. The chairman notes that similar questions have already arisen for crypto software developers and interfaces providing access to protocols. That continuity does not automatically resolve the case of an autonomous agent. It calls for distinguishing the code provider, the user, the permissions given to the software and the decisions it can actually make, rather than treating every application as the same kind of intermediary.

Market risks alongside potential benefits

Selig emphasizes that systems based on similar models could display correlated trading behavior. Their speed and unexpected responses can complicate market surveillance and risk management. He also describes beneficial uses, including more accessible analysis, detection of suspicious activity and exposure monitoring. These benefits are possibilities discussed by the regulator. They are neither an independent performance comparison nor proof that an agent can replace controls or human accountability.

Finally, the speech addresses the computing capacity, processors and electricity supporting these systems, as well as financial instruments that could hedge the associated risks. The CFTC says it has already received more than a hundred submissions on compute-related derivatives. For crypto businesses, the forum's new contribution therefore lies in considering autonomy, accountability and infrastructure together. The next development to watch is whether these discussions become formal proposals, with a clearly defined scope and adoption process.