For agricultural businesses and market intermediaries evaluating artificial intelligence (AI), the practical question is not simply whether a tool can process more information. It is whether the business can use that information reliably, protect customer data, and supervise the decisions that follow. The Commodity Futures Trading Commission’s (CFTC) staff advisory on AI provides a starting point: Using new technology, including a vendor’s system, does not displace the requirements that apply to the underlying activity.
This third and final installment in our AgCon 2026 series turns from product design to the information and tools used to make business decisions. Our first post, “Innovation in Agricultural Derivatives: AgCon Panel Examines Perpetuals, 24/7 Trading, and Tokenized Collateral,” examined whether new structures serve commercial users. Our second, “Agricultural Prediction Markets: Can Event Contracts Fill a Hedging Gap? AgCon 2026, Part 2,” applied that question to event contracts. Here, the same test applies to AI and the more advanced capabilities discussed under the label “super intelligence”: What problem does the tool solve, and what controls make its use workable?
Turning Disconnected Data Into Business Decisions
A useful starting point is a defined information problem, not a technology label. For a commercial agricultural business, that might mean testing whether a tool can connect operational records with market information to support purchasing, production, or hedging analysis. The value lies in improving a decision the business actually makes, not simply in producing more analysis.
The opportunity extends beyond trading signals. The CFTC Technology Advisory Committee (TAC) subcommittee’s 2024 report on responsible AI identifies potential uses in investment research, customer service, surveillance, risk management, and back-office processing. It also identifies risks involving unreliable outputs, sensitive information, and dependence on outside providers. These are potential applications, not assurances that a particular system will improve accuracy or reduce costs.
The first two installments distinguished an innovation’s availability from its commercial usefulness. The same distinction matters here. Firms may want to select a task, identify the data it requires, and compare results with an existing process before expanding the tool’s role. For a research application, that might mean evaluating whether the analysis is accurate and traceable to its inputs, rather than merely faster to produce.
Customer Information Is a Starting Point, Not an Afterthought
For a futures commission merchant (FCM), a central deployment question is what happens to customer information once it enters the system. A proposed efficiency gain may warrant a different assessment if the tool receives customer positions or communications rather than approved internal reference materials. Firms may want to evaluate the information involved before deciding how broadly a tool can be used.
The 2024 advisory-committee report identifies risks involving sensitive data and limited visibility into outside systems, including data inputs and retention. In evaluating those risks, firms may want to ask whether customer information is stored, reused for model training, or accessible beyond the intended users. A vendor’s description of a tool’s capabilities does not answer those separate questions about data handling.
An incremental rollout may help make those distinctions concrete. Firms may want to document permitted inputs, assign an accountable business owner, and establish who reviews outputs and escalates problems. The assessment may also address what happens if the provider changes the system or access is interrupted.
Assess the Function, Not Just the AI Label
A deployment review may benefit from asking which business function the tool will perform. CFTC Staff Advisory No. 24-17, issued December 5, 2024, describes how existing Commodity Exchange Act (CEA) and CFTC requirements may apply to regulated firms’ AI use, including systems supplied by outside providers.
For exchanges and other market infrastructure, the advisory addresses trade processing, surveillance, and system safeguards. For intermediaries and other registrants, it discusses risk management, recordkeeping, disclosures, and customer protection. Its FCM example concerns accounting for segregated customer funds, which is distinct from safeguarding customer information. A tool that summarizes research and one that supports margin calculations therefore call for different reviews, even if both use AI.
Separately, the 2024 TAC subcommittee report recommended that the CFTC consider a sector-specific risk framework informed by the National Institute of Standards and Technology’s (NIST) AI work. That recommendation is not itself an adopted commission rule, and the staff advisory is not an exhaustive compliance checklist. For a deployment decision, firms may want to distinguish applicable legal requirements from staff views, policy recommendations, and internal governance choices.
Market Integrity: Familiar Conduct, New Tools
AI presents a second question for trading and compliance teams: How does the system behave in the market? The advisory-committee report discusses both the potential for AI-enabled manipulation and the use of machine learning to detect it. For a firm evaluating a trading tool, the relevant review may extend beyond its stated strategy to its order-placement and cancellation behavior.
The distinction between technology and conduct remains important. The CEA’s spoofing provision, 7 U.S.C. § 6c(a)(5)(C), addresses orders placed on or subject to a registered entity’s rules with an intention to cancel before execution. In its 2013 interpretive guidance, the CFTC distinguished that intent from cancellations or modifications made in a genuine effort to complete a trade. It also explained that a partial fill does not automatically exclude a spoofing violation. The inquiry concerns conduct, intent, and surrounding circumstances, not whether the system is described as AI, super intelligence, or conventional automation.
The technology may also support detection. Staff Advisory No. 24-17 discusses potential AI use by exchanges and swap execution facilities to identify abusive activity and unusual trading patterns, while emphasizing sufficient compliance staffing and resources. Firms considering surveillance tools may want to assess not only whether a system generates alerts, but whether reviewers can understand them, investigate the activity, and document the disposition. More alerts do not, by themselves, demonstrate more effective oversight.
Match Oversight to the Decisions the Tool Can Make
Human oversight is not a single control. The 2024 report distinguishes different human roles in automated systems and recognizes that real-time review may be impractical for high-speed activity. Firms may want to identify where a person can review an output, intervene, or authorize a broader use. Naming an overseer is different from giving that person the information, training, and authority to perform the role.
For a proposed application, useful questions include: Does the tool only provide information, or can it initiate an action? Who approves changes to its role? What information is available to investigate an unexpected result? How are decisions documented, and can the decisions be functionally audited? Firms may want to resolve those questions before moving from internal assistance to customer-facing or trading functions. That approach links oversight to the consequences of an error rather than to the sophistication of the product’s label.
Across all three installments, the common test is whether an innovation helps agricultural businesses manage an identifiable exposure or operational problem. For derivatives, that means examining contract terms, usable liquidity, and the connection to physical markets. For AI, it means examining information inputs, output reliability, and responsibility for resulting decisions. Neither a new product structure nor a more capable tool establishes commercial value on its own.
Practical Takeaways
- FCMs and other intermediaries may want to consider mapping customer-data flows before approving a new tool, including access, retention, training use, and vendor arrangements. Internal assistance may warrant a different assessment from applications affecting customer communications, funds, or execution.
- Commercial firms and producers may want to consider starting with a defined operational problem and testing whether the tool makes existing information more actionable. Performance measures may help assess whether to extend its role across the business or supply chain.
- Trading and compliance teams may want to consider testing order-placement and cancellation behavior, evaluating surveillance alerts, and documenting review and escalation. Technical training may help reviewers understand when and how to intervene.
- Legal and risk leaders may want to consider separating applicable legal obligations from nonbinding staff guidance and policy recommendations when assessing a deployment. Continued attention to CFTC engagement may help identify where expectations are developing.
Market participants evaluating AI tools may benefit from connecting the proposed business use to its regulatory and supervisory implications.
Contact Us
Husch Blackwell’s Securities & Commodities Regulatory & Enforcement team advises trading firms, end users, and financial institutions on derivatives compliance, governance, and supervisory frameworks, and represents clients in market-conduct investigations. Contact Jeff Le Riche or Kip Randall to discuss how those services may support evaluation of an AI deployment and the controls surrounding its use.