BNY and Microsoft complete first agentic negotiation of a US Treasury bill trade
For the head of markets at a US dealer bank
BNY and Microsoft completed a pilot trade negotiation of a U.S. Treasury bill using AI agents, BNY said in a press release dated Nov. 17.1
BNY built the Agentic Negotiation Protocol (ANP), a control layer that lets authorized agents from different organizations negotiate securely, according to the press release.1 BNY said the approach streamlines the traditional message-based negotiation process and frees traders for other work, while keeping human accountability and existing risk and transaction-processing controls.1
Microsoft Treasury gave its agent specific trading instructions, and the agent connected to BNY through a secure agent gateway to ANP, BNY said.1
BNY controls validated the counterparty, security, size, settlement terms and available pricing, then the agents negotiated prices within their authorized limits, according to BNY.1
The agreed terms were independently submitted for comparison against ANP's central ledger to confirm both parties had the same understanding, and BNY and Microsoft traders then approved them, with booking done through existing systems, BNY said.1
Agents cannot execute the trade for either firm, and ANP keeps a record of participants, messages, control checks and human decisions so the negotiation can be reconstructed, according to the press release. The release did not give the trade size, price or date.1
Human sign-off stays in place, so the gain BNY describes is fewer back-and-forth messages and a shared record of what was agreed. The record sits in a ledger BNY built, and the trade was a single bilateral Treasury bill, so banks should treat it as a pilot of the control design, not proof of scale.
BNY plans to keep testing ANP, expand beyond the initial bilateral use case and evaluate other negotiated workflows, and said future uses could include repo and other contractual negotiations.1
Our take. BNY wrote the rulebook for agent-to-agent Treasury negotiation, filed for a patent, and runs the ledger that checks both sides agree. Join its protocol or build your own audit record before a counterparty's agent asks which you use.
Action. Markets compliance: Map how trader approvals of agent-negotiated terms would be recorded and reconstructed under your current trade surveillance and recordkeeping rules.
What our records say
BNY runs its AI strategy through one enterprise platform, Eliza, on which staff build agents and which supports "digital employees"; CEO Robin Vince put AI solutions in production at "400-plus" on the Q3 2026 earnings call (October 15, 2026).1 The bank counted 117 AI solutions in production at September 30, 2025, about 220 (alongside 140 'digital employees') in April 2026 and '400-plus' by Vince's count on the October 15, 2026 earnings call.1
From our records
- BNY (Bank of New York Mellon)bank · latest Microsoft customer story: Copilot built into Eliza; Work IQ and Copilot Cowork next (undated page) (Nov 2026) · vendors Quantexa, Snapdocs, Microsoft Copilot
- U.S. Treasuryregulator · binding Voluntary framework · latest FSOC AI Innovation Series (Mar–May 2026) (Jun 2026), Treasury FS AI RMF and AI Lexicon (Feb 2026) (Feb 2026)
Sources
- Primary sourceBNY and Microsoft Complete First Agentic Negotiation of a U.S. Treasury Trade Bny
- Also reported byAI agents in bond market arrive with a stealthy Microsoft trade Businesstimes