Huang: Embed machine-speed rules in AI-to-AI commerce

Sydney Huang, chief executive of Human API, warned that agentic AI executing transactions without human intervention could produce inflation spikes or market flash crashes faster than regulators can respond. Her remarks referenced an April 2026 IMF report describing a shift from “click-to-pay” to “decide-to-pay,” where autonomous agents carry out payments and trades.

Huang said agent-to-agent commerce will raise the velocity of money by reducing human delays. She estimated machine-to-machine flows could be up to ten times faster than today. Central bank tools such as interest-rate changes work through human behavior over months; Huang noted those lags would shrink or disappear in an agentic marketplace.

To address that timing gap, Huang called for controls that run at machine speed. She recommended real-time monitoring systems, programmable compliance embedded in financial infrastructure, and automated transaction-level circuit breakers to stop cascading failures. Those measures reflect elements of the IMF’s Three-Layer Framework, which requires the authorization layer of transactions to carry human-defined mandates.

Huang said regulators will need to express rules in machine-readable formats so agents can enforce them at the point of transaction. She proposed automated “fuses” that break chains of highly correlated agent behavior, and tighter integration of know-your-customer and anti-money-laundering checks into agent logic.

Detecting wrongdoing in a market where agents rarely use human language will rely on behavior analysis rather than message inspection, Huang noted. Regulators would look for synchronized actions, shared data dependencies and statistical anomalies. She proposed “decision provenance,” a requirement that agents produce verifiable records showing a decision followed an independent, declared policy. That record would help distinguish lawful optimization from coordinated pricing behavior.

Huang said common standards are needed for safe agent negotiations across organizations. She pointed to emerging protocols such as the agent payments protocol (AP2) and the model context protocol (MCP) as tools that let one company’s agent verify another’s identity and authorization, operate under shared negotiation rules, and attach verifiable guarantees to actions. Those protocols shift trust from single counterparties to system-level guarantees enforced in code.

She also raised operational concerns about long-term reliance on autonomous agents. If corporate treasuries and trading desks are run by bots for years, human staff may lose the skills required to intervene in a crisis. Huang urged regular drills where humans retake control, simulation modes that mirror agent logic for comparison, and practiced pathways for kill switches. “Maintaining operational readiness is as important as building fallback mechanisms,” she said. She added that the kill-switch procedure must be practiced, not only designed.

Huang cited a projected $236 billion agentic market by 2034 and said regulators and firms should redesign financial architecture and law to operate at machine speed. She warned that if human-defined mandates are not embedded at the architectural level, transaction systems could outpace the ability of creators and regulators to manage them.