Sydney Huang, an economist tracking financial system risks, has warned that AI bot collusion could spread across markets before regulators have time to respond. The threat, Huang says, stems from the growing speed of AI-to-AI commerce — machines trading directly with other machines in milliseconds. That velocity, Huang argues, could leave central banks unable to react to inflation spikes or flash crashes that unfold faster than humans can even detect.
The speed problem in machine-run markets
Huang's warning relies on an April 2026 International Monetary Fund report that describes the end of what economists call policy lag — the natural delay between a problem emerging and a central bank tightening or loosening money. In an AI-dominated system, that lag vanishes. A bot can trigger a cascade of sell orders across thousands of algorithms before a human policymaker finishes reading the morning briefing. Huang points out that this isn't a hypothetical future; some high-frequency trading firms already operate entirely through AI agents that negotiate prices with each other without human oversight.
Why regulators can't keep pace
The core issue, according to Huang, is that regulators still think in terms of quarterly reports and monthly data releases. When machines can execute a coordinated strategy — deliberately or accidentally mimicking collusion — in seconds, waiting a month to spot the pattern means the damage is already done. Huang specifically warns that AI bots could learn to signal each other through pricing patterns, effectively colluding without explicit communication. That kind of behavior is nearly invisible to current surveillance tools.
Embedding rules directly into code
Huang isn't just raising alarms. The economist suggests a concrete fix: build regulatory frameworks directly into the AI code that runs these trading systems. Instead of trying to police behavior after the fact, central banks and market overseers could require that all trading algorithms contain hard-coded constraints — speed limits, position caps, circuit breakers that trigger automatically. The IMF report cited by Huang argues that such embedded rules are the only way to prevent cascading failures when machines trade at machine speed. But enforcement would require global coordination, which has been historically slow.
Whether regulators will move fast enough to embed those rules before the next AI-driven crash isn't clear. Huang notes that the window for action is narrowing, and the IMF's April 2026 deadline for its recommendations adds urgency. For now, the machines are already trading.
