Federal agencies and market groups outline AI risks and tools at CFTC 'AI Day'

The Commodity Futures Trading Commission Technology Advisory Committee on AI Day heard agency and industry experts describe how they are experimenting with artificial intelligence while attempting to manage its market risks.

Federal Reserve Chief Innovation Officer Sunena Tuta described a three-part institutional approach—education, enablement and experimentation—centered on an in-house sandbox called Launchpad. The Fed’s Launchpad offers quarantined, cloud-native environments where staff can test generative AI capabilities with guardrails before any production deployment, she said. "We unlocked as many of the generative AI capabilities but we did it in a guarded environment," Tuta said, adding that the Fed frames AI as a “co-pilot” for humans rather than a replacement: "It's for this technology to be your co-pilot but it's nowhere near to kind of fly the plane on autopilot."

Kirsten Wagner of the Modern Markets Initiative traced lessons from two decades of market automation: algorithmic trading reduced costs and improved market efficiency, but it also generated concentrated data advantages, litigation over data access and episodes of market disruption. Wagner said governance, explainability and third-party vendor scrutiny are central to preserving fair access and market integrity as AI expands in trading.

Elham Tabassi of the National Institute of Standards and Technology summarized the NIST AI Risk Management Framework (AI RMF), which the agency designed to be flexible, measurable and context-specific. NIST emphasizes a socio-technical approach and published companion materials on generative AI risks and synthetic-content authentication. "If we cannot measure it we cannot improve it," Tabassi said, describing NIST's emphasis on measurable trustworthiness attributes.

Todd Conlin of the Treasury Department reviewed a publicly available Treasury report on AI-specific cyber and fraud risks, based on interviews with 42 sector participants. Conlin said firms widely use AI for cybersecurity and fraud detection but that smaller institutions face capability and data gaps. He highlighted growing concerns about AI-enabled voice and video mimicry being used in targeted fraud against high-value customers.

Across presentations, panelists and TAC members repeatedly urged practical governance: embed legal, compliance and cyber teams early in experiments, document transparency and testing practices, and build mechanisms for industry information sharing on fraud signals. Speakers also called for a risk-based, principles-first approach rather than prescriptive technical rules, with NIST identified repeatedly as a shared reference point.

The presentations preceded a subcommittee report adoption later in the meeting; that report recommends the CFTC host outreach, consider the NIST AI RMF for registered entities, and build staff capacity to implement AI-related oversight.