BNY and Microsoft have completed what BNY calls an industry-first pilot in which AI agents negotiated a U.S. Treasury bill trade, according to a BNY press release on 17 November. Bloomberg first reported the trade that morning. Human traders at both firms reviewed and approved the terms, and the agents cannot execute a trade, BNY said.
Our count from BNY’s description: BNY’s controls validated five items before the agents could negotiate price, the agreed terms were matched once against a central ledger, and traders at both firms signed off. The bank says it oversees $63.4 trillion in assets under custody and/or administration.
Business Pill · Least Privilege
An evergreen Business Engineer explainer on giving an AI agent only the access its task needs. It teaches the general idea only and does not discuss BNY or Microsoft.
The key insight: BNY and Microsoft had AI agents negotiate a U.S. Treasury bill trade inside BNY’s Agentic Negotiation Protocol, BNY said; traders at both firms approved the terms and the agents cannot execute a trade.
How the Trade Worked
Microsoft Treasury gave its agent specific trading instructions, and the agent connected to BNY through a secure agent gateway to BNY’s Agentic Negotiation Protocol, or ANP, the release says. BNY’s controls validated “the counterparty, security, size, settlement terms and available pricing”. Once those checks passed, the agents negotiated prices within their authorized limits.
When the agents reached agreement, the proposed terms were submitted for comparison against ANP’s central ledger. “BNY and Microsoft traders then reviewed and approved the proposed terms, with execution and booking completed through the firms’ existing systems and controls”, BNY said.

The Agents Do Not Execute
BNY’s release is explicit on the limit: “Agents cannot execute the trade on behalf of either firm and trades are booked through existing internal trading systems”. ANP keeps a record of who participated, the messages exchanged, control checks and human decisions, so the negotiation can be reviewed and reconstructed, according to the release.
That is a different design from the test JPMorgan disclosed the same day. JPMorgan and Hyundai Card said AI agents carried out a simulated foreign exchange transaction without direct human intervention, CNBC reported. BNY’s pilot was a Treasury bill trade with human approval at the end.
Where BNY Says It Goes Next
BNY plans to keep testing ANP, beyond the initial two-party use case, and says future applications could include repo and other contractual negotiations. BNY’s Laide Majiyagbe said “high-touch” trading does not go away and sees a gap between voice-negotiated and fully electronic trading that agents can fill, according to Investing.com’s summary of Bloomberg’s report. She said smaller players could access negotiated trading where hiring dedicated traders “didn’t make sense”.
The Business Lever
The lever is the control layer, not the agent. In BNY’s description, ANP verifies participants, checks terms against BNY’s own controls and keeps the audit record, while each firm’s agent works inside limits its owner sets. Companies are setting limits on their own staff’s AI use too: Salesforce caps AI tokens at $1,000 a month. Microsoft is also selling models to enterprises, including MAI Code 2.
The Structural Read
As we read BNY’s release, the product is the control layer: ANP verifies the agents, checks the counterparty, security, size, settlement terms and pricing, and keeps a record a reviewer can reconstruct.
As we read the two disclosures of 17 November, BNY kept a human approval and execution in existing systems, while JPMorgan and Hyundai Card described a simulated transaction without direct human intervention, CNBC reported.
BNY press release, 17 November 2026
“Agents cannot execute the trade on behalf of either firm and trades are booked through existing internal trading systems”
Three Implications
FOR TREASURY DESKS Microsoft Treasury gave its agent specific trading instructions and connected through a secure agent gateway, BNY said.
FOR BNY BNY says it oversees $63.4 trillion in assets under custody and/or administration and plans to keep testing ANP beyond the first two-party use case.
FOR BOND TRADING Future applications could include repo and other contractual negotiations, the release says.
The Business Engineer Lens
This story maps onto the Business Engineer framework AI & The Harness Theory.
The framework says: “It is an operating system that wraps your AI — a harness that compresses your thinking, your team’s feedback, and the model’s raw capability into something replicable.”
As we read it, ANP’s rules and checks are a harness of this kind around the negotiating agents. The framework predates this article and does not discuss BNY or Microsoft.
What Is Not Established
BNY did not disclose the size, price or maturity of the Treasury bill, or when the pilot trade took place. The release does not say which AI models the agents used. BNY says ANP is patent pending. We did not contact BNY or Microsoft.
The Bottom Line
BNY and Microsoft ran a pilot in which AI agents negotiated a U.S. Treasury bill trade under BNY’s ANP framework, with human traders at both firms approving the terms and execution kept in the firms’ existing systems. BNY plans further tests, including repo.
97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. The pilot, how the agents negotiated, the controls, the trader approvals and BNY’s next steps are as described in BNY’s 17 November 2026 press release, which we read in full. Bloomberg reported the trade first, on 17 November; Laide Majiyagbe’s remarks are as summarised by Investing.com from Bloomberg’s report. The count of control steps is ours, from BNY’s description. We first saw the story via a relay post on X on 18 November. We did not contact BNY or Microsoft.
Sources: BNY press release, 17 Nov 2026 · Bloomberg, 17 Nov 2026 · Investing.com, 17 Nov 2026









