BNY and Microsoft complete Treasury trade negotiation
Pilot shows how autonomous AI agents could transform capital markets negotiations under supervision.

BNY and Microsoft have completed the first agentic negotiation of a US Treasury trade.
It comes as financial institutions race to deploy generative AI and autonomous agents into capital markets workflows.
BNY said the pilot demonstrates how AI agents can negotiate transactions within strict controls while keeping humans firmly in charge of trading decisions. “This milestone demonstrates how AI agents can be used responsibly in institutional markets, guided by clear deterministic rules and guardrails and subject to internal review processes,” the company explained.
The trade was negotiated using BNY’s Agentic Negotiation Protocol (ANP), a framework designed to allow AI agents from different organisations to securely exchange offers and counteroffers while operating within predefined limits.
In the pilot, Microsoft Treasury instructed its AI agent to negotiate a US Treasury bill trade. The agent connected to BNY through a secure gateway, where both firms’ systems validated counterparties, settlement terms and pricing parameters before negotiations began. Once agreement was reached, human traders at both organisations reviewed and approved the transaction before execution.
According to BNY, the approach streamlines traditional message-based negotiation processes, reducing manual workload while preserving accountability, auditability and existing risk controls. The protocol also records all messages, control checks and human decisions, creating a transparent record that can be reviewed after the event.
The initiative reflects growing interest in agentic AI across financial services, where firms are increasingly exploring autonomous systems capable of carrying out complex tasks under human supervision. While many institutions have focused on internal productivity use cases, the BNY-Microsoft pilot demonstrates how AI agents could begin interacting directly with one another to support market operations.
BNY plans to continue testing the technology beyond bilateral Treasury trading, with future applications potentially spanning repo markets and other negotiated financial contracts.
The development marks a significant step towards agent-to-agent commerce in capital markets, suggesting how AI could reshape institutional trading workflows while maintaining the governance standards required by heavily regulated financial markets.



