CFTC floats a 'know your agent' rule for AI trading in markets: the key question is who is accountable

By CriptoNoticias · October 28, 2026.

The US Commodity Futures Trading Commission (CFTC) is examining what legal framework is needed for artificial intelligence agents that operate in markets with little human intervention. Its chairman, Mike Selig, raised the issue at a forum held on Wednesday, according to CriptoNoticias.

Selig put three questions on the table without presenting a finished proposal. First, who is liable if an agent executes fraudulent trades. Second, whether the developers of these systems should register with the regulator. Third, whether a specific identification mechanism is needed, under the concept of "know your agent" (KYA). KYA would be a version of "know your customer" (KYC), the process financial intermediaries use to verify the identity of whoever is trading.

According to the article, Selig noted that the experience with crypto assets could serve as a reference. The sector has already had to work out when developing software or interfaces triggers obligations typical of a traditional financial intermediary. To illustrate the scope, he cited a household appliance that buys groceries on its own with funds earned by trading cryptocurrencies.

At the same forum, Brett Harrison, CEO of derivatives trading firm Architect, said that models such as ChatGPT, Claude and Gemini make it possible to analyze news and other data to generate trading decisions. According to Harrison, this broadens access to algorithmic strategies for large institutions and retail investors alike. The CFTC also announced a roundtable on 24/7 trading for December 15.

Our take: what matters is not the acronym but the shift in the question. Until now, financial regulation identified the person or entity behind each trade. With autonomous agents, that chain gets longer: user, agent, developer, platform. If no one is clearly obliged to answer for it, fraud becomes cheaper to commit and harder to prosecute.

KYA is a reasonable path, but it has limits worth anticipating. An agent registry only works if it verifiably identifies the ultimate responsible party, not a piece of software anyone can clone. Requiring every developer to register could also penalize open-source software and small teams, precisely where much of the innovation happens. As industry context, this approach ties in with the trend in agent governance toward verifiable credentials and auditable records, rather than relying on internal policies.

For now, these are a regulator's open questions, not rules or a formal proposal. We advocate evidence-based regulation focused on traceability of who decides and who is accountable, not panic over the technology. In the short term there will be friction and legal gaps. In the long term, markets with well-identified and supervised agents could open up access to tools currently reserved for large institutions, provided accountability is not diluted along the way.