The U.S. Commodity Futures Trading Commission (CFTC) is beginning to confront a regulatory reality that many in crypto and traditional finance have been discussing for months: software agents may soon do more than recommend trades—they may autonomously hold assets, execute transactions, and potentially interact with digital markets.

CFTC Chair Michael Selig made that case in remarks delivered this week at the CFTC’s inaugural Frontier Forum in Washington, D.C., arguing that regulators can’t assume human beings remain the central “market participants” behind every action taken in markets. He described autonomous AI agents that could place trades on behalf of users—raising questions about registration, liability, and how regulators can tell whose interests are being served.

Key takeaways

CFTC Chair Michael Selig said autonomous AI agents that can hold assets and execute trades are “no longer science fiction,” pushing regulators to rethink existing frameworks.
Selig questioned how current rules—built around identifiable human participants—should apply when software acts on a user’s behalf.
He highlighted open issues including developer registration, responsibility if an agent malfunctions, and how regulators should track which agent is trading for which party.
Selig argued AI could improve risk management for smaller traders, but autonomy also introduces risks such as correlated behavior and faster market reactions.
Exchange product momentum already suggests agent-driven trading workflows are arriving, with Coinbase and Binance rolling out agent-related tools earlier this year.

Why the CFTC is focusing on autonomous agents now

Speaking at the CFTC Frontier Forum, Selig positioned the coming shift as more than incremental automation. He referenced President Trump’s framing of “super intelligence” and suggested the technology is now as commercially tangible as historic industrial breakthroughs such as the printing press, the steam engine, and the internet.

That framing matters because the CFTC’s regulatory approach has traditionally been built around identifiable market participants conducting defined functions. In Selig’s view, autonomous software complicates that assumption: if an AI agent can identify trading opportunities and execute transactions for a user, the relationship among user, developer, agent, and market activity becomes harder to map.

His remarks pointed to practical enforcement dilemmas. For example, if the “developer” of an autonomous agent is not the person trading directly, regulators still need to determine who falls within existing registration and compliance expectations. Selig also raised the question of liability when an agent behaves unexpectedly—whether through a bug, a model failure, or an unforeseen reaction to fast-moving markets.

Perhaps most importantly for market oversight, he asked how regulators can know which agent—or set of agents—is transacting on whose behalf. He suggested that today’s compliance tools may need an update, floating the idea of a “know-your-agent” concept alongside or in place of more familiar know-your-customer expectations.

Potential benefits: faster processing and broader access to risk tools

Selig argued that agent-driven finance could deliver real advantages, particularly for smaller participants that can’t replicate institutional infrastructure. He said AI could process large volumes of information more quickly than humans and potentially with greater accuracy—an ability that could support risk management and improve decision-making speed.

In this scenario, the “automation layer” is not just about placing orders. It could also help monitor conditions, react to signals, and manage exposure in ways that previously depended on specialized teams, expensive tooling, or extensive market experience. That idea aligns with broader industry patterns: as consumers increasingly use AI tools to assist with investment decisions, the compliance burden shifts from advice alone to execution as autonomy rises.

Supporting that adoption trend, an eToro survey published in September found that 79% of U.S. millennial investors and 77% of Gen Z investors already used or were open to using AI to select or adjust investments. While the survey doesn’t measure how many users employ fully autonomous trading, it underscores how quickly AI-enabled finance may become mainstream—meaning regulators are likely to face these questions sooner rather than later.

New risks: correlated behavior, unexpected reactions, and tougher surveillance

Alongside potential benefits, Selig emphasized that greater autonomy can create new forms of risk. He described scenarios in which agents may react in unexpected ways to changing market conditions. This isn’t purely hypothetical: if multiple traders or institutions rely on similar models, strategies, or underlying data sources, there is a risk of correlated trading—where many independent agents behave similarly and amplify volatility.

Speed is another issue. Autonomous transactions can occur quickly enough to outpace conventional monitoring workflows, making market surveillance and risk management more complex. For regulators, the challenge becomes not only detecting misconduct, but also distinguishing between legitimate agent behavior and harmful system dynamics that arise unintentionally.

The problem of shared inputs and shared outputs has also been raised in the industry. Animoca Brands chairman Yat Siu recently told Cointelegraph that if a large group of people asks AI about the same trade, the answers could converge—especially when users start from similar prompts. He added that users’ own knowledge and follow-up questions can affect personalization, but the core risk remains: common models and common assumptions can yield common behavior.

Regulators outside the U.S. have made similar concerns. Last year, the Bank of England’s Financial Policy Committee warned that widespread use of similar models could amplify market shocks, reflecting a growing recognition that model homogeneity can be a systemic issue, not just a technical one.

Agent trading products are already emerging

Selig’s remarks land at a time when crypto exchanges are building tools that make agent-driven workflows more practical. That product momentum suggests the “how do regulators oversee this?” debate is no longer theoretical.

Coinbase, for instance, introduced “Agentic Wallets” in February, describing a setup where AI agents can spend and trade within user-defined limits. Binance has also moved in this direction: it launched “AI Agent Skills” in March, positioned as a way for agents to access market data and place or manage spot orders.

These features don’t resolve the regulatory questions Selig raised, but they do illustrate a trend: execution can increasingly be delegated to software, and user intent may be translated into transactions without continuous manual oversight. That shift affects how compliance responsibilities may need to be structured—especially around authentication of counterparties, audit trails, and clarity of who ultimately controls agent behavior.

What to watch next

As agentic finance expands from “assisted decisions” to “autonomous execution,” the next regulatory pressure points will likely be practical: defining who is responsible for agent behavior, how to trace the link between agents and users, and whether compliance frameworks need a parallel “know-your-agent” lens. For investors and builders, the key question is whether the rules evolve in step with the capabilities of deployed systems—before autonomy becomes the default mode of interacting with markets.