Sydney Huang, CEO of Human API, has issued a stark warning about the growing risk of collusion among artificial intelligence agents in financial markets. As the world transitions to a machine-speed economy, she argues that regulators must embed oversight directly into code to prevent runaway systemic failures.
AI Agent Economy: Money Velocity Could Surge 10x
According to an April 2026 report from the International Monetary Fund (IMF), the global economy is shifting from a 'click-to-pay' to a 'decide-to-pay' era. As humans leave the transaction loop, autonomous AI agents will dramatically increase the velocity of money. Huang predicts a potential 10x increase in money circulation speed. This poses a nightmare for central banks, which rely on policy 'lag'—the months it takes for rate changes to filter through the human system. In an AI-to-AI economy, that lag disappears, leaving regulators helpless against machine-speed inflation spikes or sudden crashes.
Regulation Must Be Embedded in Code: Real-Time Monitoring and Automatic Circuit Breakers
Huang insists that regulators must stop being spectators and become part of the code itself. She advocates for real-time monitoring systems, programmable compliance baked directly into financial infrastructure, and automatic circuit breakers to halt cascading failures. This aligns with the IMF's proposed 'Three-Layer Framework,' which mandates that each transaction's authorization layer contain human-defined, embedded instructions. Policies must also be expressed in machine-readable formats for enforcement at the transaction level. When agents exhibit highly correlated behavior, autonomous 'fuses' should trigger to stop chain reactions.
Invisible Markets and Decision Provenance
A key challenge is the 'invisible market' where agents coordinate without human language. How to distinguish legitimate optimization from collusive price-fixing? Huang calls for a shift from communication analysis to behavior analysis—examining synchronized actions, shared data dependencies, and statistical anomalies. The solution lies in 'decision provenance': requiring agents to provide verifiable proof that decisions were made independently according to declared policy. Additionally, agents need universal standards for identity, communication, and enforcement, such as the Agent Payment Protocol (AP2) and Model Context Protocol (MCP), allowing seamless and secure cross-enterprise transactions without proprietary intermediaries.
Human Skill Atrophy and Contingency Preparedness
As governance is increasingly delegated to digital proxies, human operators risk skill atrophy. Huang warns that a treasury manager who hasn't intervened for five years may be incapable of handling a crisis. She urges regular human-in-the-loop drills where operators simulate agent actions to compare logic, and practice 'kill switch' procedures. 'The goal is to ensure human oversight remains functional and practiced, not theoretical,' she said.
Analytics firm Nansen predicts AI agents will dominate crypto investing by 2028. Huang concludes: 'To govern a machine-speed economy, the law must become as fast as the machines. If we fail to embed the human-in-the-loop at the architectural level, we risk building an economy that moves too fast for its creators to control.'