
Key Points
- 01BNY and Microsoft ran a pilot U.S. Treasury bill trade negotiated by AI agents
- 02BNY’s Agentic Negotiation Protocol governed the autonomous price talks
- 03Human traders at both firms retained final approval and executed the deal
- 04BNY aims to extend the protocol to more transaction types and negotiated workflows
Industry-first AI-negotiated Treasury bill trade
BNY and Microsoft have completed an industry-first pilot in which artificial-intelligence agents negotiated the terms of a U.S. Treasury bill trade. The agents bargained over price and reached agreement before human traders at both firms reviewed and approved the transaction, which was then executed and booked through existing trading and risk-control systems.
The pilot is described as the first U.S. Treasury trade negotiated by software agents in the government bond market. Neither company has disclosed the size, maturity or price of the Treasury bill involved, or the precise time at which the trade was executed.
How BNY’s Agentic Negotiation Protocol works
At the center of the pilot was BNY’s Agentic Negotiation Protocol (ANP), a framework and control layer that allows authorized AI agents from different organizations to negotiate securely and transparently. ANP verifies participants and connects them through a secure gateway, enabling agents to exchange offers and counteroffers and to confirm a shared understanding of proposed terms.
Participant-side controls are designed to provide guardrails through predefined rules and AI safety checks, keeping agents within tightly defined parameters based on approved pricing inputs. Before seeking trader approval, ANP’s ledger checks that both parties’ agents have submitted matching terms and maintains a record of all messages, control checks and human decisions so that the negotiation can be reviewed and reconstructed.
In the pilot, Microsoft’s treasury department gave its agent specific trading instructions describing acceptable deal boundaries. BNY’s controls validated the counterparty, security, size, settlement terms and available pricing, after which the agents negotiated prices within their authorized limits under ANP’s governance.
Maintaining human authority and market controls
The design of the system keeps human traders in control of final decisions and actions. The AI agents are not permitted to execute trades on behalf of either firm, and all bookings occur through existing internal trading and transaction-processing systems, preserving established risk and compliance controls.
The approach is intended to streamline the traditional, message-based negotiation process by introducing more consistency and allowing traders to focus on higher-value work, while maintaining human accountability. The protocol preserves an auditable record of who participated, what was communicated, and how decisions were made.
Future testing and broader applications
BNY plans to continue testing the Agentic Negotiation Protocol beyond the initial bilateral Treasury bill use case. The firm is evaluating additional negotiated workflows and considering applications for other transaction types such as repurchase agreements and other contractual negotiations.
Plans also include extending the protocol to other negotiated workflows, including repurchase agreements, foreign-exchange payments, collateral terms and broader contractual negotiations. Such tools are described as potentially lowering the barrier to entry for market participants for whom a dedicated human trading effort may not be economical.
Key Takeaways
- 01BNY and Microsoft have demonstrated that AI agents can negotiate a live U.S. Treasury bill trade within strict guardrails while leaving final authority to human traders.
- 02BNY’s Agentic Negotiation Protocol functions as a secure control layer, combining identity verification, safety checks and an auditable ledger for agent-to-agent negotiations.
- 03The pilot focuses on improving efficiency in negotiated trading without removing human oversight, aiming to streamline workflows while preserving existing risk controls.
- 04BNY intends to build on this proof of concept by expanding AI-negotiated workflows to more transaction types over time.
- 05The size and price of the pilot trade remain undisclosed, underscoring that the current focus is on validating the negotiation framework rather than showcasing trade specifics.
References
- https://finance.yahoo.com/technology/ai/articles/bny-microsoft-complete-first-ai-201911277.html?.tsrc=rss
- https://www.bizjournals.com/pittsburgh/news/2026/11/17/bny-microsoft-ai-treasury.html
- https://www.marketscreener.com/news/bank-of-new-york-mellon-bny-and-microsoft-complete-first-agentic-negotiation-of-a-u-s-treasury-tr-ce785edcdb81f625
- https://leprivatebanker.com/2026/11/17/bny-and-microsoft-have-ai-agents-haggle-over-a-treasury-bill-trade-with-humans-signing-off-on-the-price/