BNY and Microsoft Use AI Agents to Negotiate a Treasury Bill Trade By Matthew Brienen Published: 2026-11-17T16:10:18.831+00:00 BNY and Microsoft completed a pilot in which AI agents negotiated the terms of a U.S. Treasury bill trade. Human traders reviewed and approved the proposed terms before execution, according to BNY. How the pilot worked The companies used BNY’s Agentic Negotiation Protocol, or ANP, to allow agents developed by BNY and Microsoft Frontier Company to negotiate across organizational boundaries. BNY described the framework as operating within defined rules and guardrails, with an auditable record and existing institutional controls. The company called the transaction an industry-first pilot trade negotiation of a U.S. Treasury bill using AI agents. That is the companies’ characterization of the milestone. The announcement does not establish a broad commercial rollout or show that AI has replaced human traders. Humans kept the final decision Negotiation is one part of a trade. The agents proposed terms within their authorized parameters; human traders retained the authority to review and approve the outcome. BNY said trades are booked through existing systems and controls. That distinction is important when assessing what the pilot demonstrates. It shows software agents can participate in a controlled negotiation between organizations. It does not mean an AI system independently decided to trade, bypassed human approval or settled the transaction on a blockchain. Why Treasury markets are a useful test U.S. Treasury bills are short-term government securities. A trade in one still involves counterparties agreeing on terms and processing the transaction through financial infrastructure. Automating part of that conversation could eventually reduce manual work, provided firms can establish reliable permissions, controls and records. The safeguards matter as much as the speed of an agent’s responses. Institutions need to know which agent is authorized to act, what limits apply, who approved the result and how disputes or errors can be reconstructed. A pilot is a way to test those questions in a narrower setting before expanding the process. What comes next BNY said further testing will look beyond bilateral negotiations and evaluate other negotiated workflows. The announcement does not provide enough detail to judge how quickly those applications will reach production or what savings they might deliver. For readers following the overlap between AI and financial infrastructure, the useful detail is the division of responsibility: agents handle part of the negotiation, while people and established systems remain responsible for approval and booking. Whether that model scales will depend on how it performs under more complex conditions and how other firms choose to participate.