Microsoft and BNY Agents Negotiate a U.S. Treasury Trade

Bank of New York Mellon and Microsoft used AI agents to negotiate a U.S. Treasury trade, which the companies describe as a first. Microsoft set boundaries for the price its agent could accept, while BNY’s Agentic Negotiation Protocol let authorized agents negotiate and confirm their understanding of the terms. The agents agreed on a price, but people approved the transaction. BNY plans to test the system in other negotiated activities, including repo deals, foreign-exchange payments, collateral terms and contracts; the agents do not yet fully match human traders.

Apple is testing macOS 27.3, now in beta 2, with tighter controls over apps’ access to sensitive data and other apps. The update splits Device Control and Data Access into System & App Protection Override and Accessibility, and lets users limit some permissions to selected folders or apps. Users can choose “Ask First” or “At Any Time” for certain control requests, while a menu-bar indicator shows when one app controls another. Existing permissions remain active, and developers are expected to ensure their apps work with limited access.

Several other products and deployments put AI to practical use. Acemate’s new AI Shot Highlights feature for its iOS and Android tennis app combines phone video with shot data, skeletal tracking, and ball-speed and placement overlays; it also improves video quality, loading times and the interface. Ridge Security says its RidgeGen platform pairs commercial language models with its security expertise, using verification checks, guardrails and reviewable logs and evidence. Automated remediation is planned for early next year. A separate review of agent deployments across banks, utilities, manufacturers and retailers finds that people, processes and data matter alongside technology.

Liquid AI released d1.2 and d1.2-ultra, which accept text, images and audio and add five answer types. The company says d1.2-ultra scored 75 on the Artificial Analysis Decision Index v2.1, the highest score among decision models listed there. Meanwhile, OpenAI CEO Sam Altman warned at a November 16 Semafor event that chip suppliers may be counting on the same future buyers, including OpenAI and Anthropic. He said those companies face limits in demand and data-center capacity, while stressing that demand for data centers remains solid and computing needs should grow over time. His concern is whether supply and demand will align in the short term.

Docusign is bringing contract work into ChatGPT, letting sales teams prepare renewal briefs that combine signed-agreement terms, deadlines and commitments with customer information from a connected CRM such as Salesforce. Sellers can route legal requests through existing Docusign workflows or create agreements from customer data using approved templates. Docusign says its MCP server uses enterprise permissions, end-to-end encryption and configurable guardrails, including for agreement creation and signature collection.

INFevo announced November 16, 2026, that it raised more than RMB 100 million in Pre-A funding to expand AI virtual-cell models and automated experiments. Xinglian Capital, Prosperity7 Ventures, Huagai Capital, Huakong Fund and DT Capital joined the round; Shunwei Capital, HSG and Songhe Capital invested again. The company aims to predict cellular responses to drug or genetic changes and test predictions in automated experiments. Its Popper Project uses research agents and databases to develop and review hypotheses, while its Karl Workbench is in beta. INFevo is also working with a Beijing key laboratory led by Peking University First Hospital on liver, biliary and pancreatic tumor organoids.

Pepperdine University’s Davenport Institute will hold a two-day workshop, “Applied AI in Local Government: Practical Skills for Public Sector Leaders,” in Malibu on December 2–3, 2026. Applications are due November 20.

Key Takeaways

  • BNY and Microsoft’s AI agents negotiated a U.S. Treasury trade under Microsoft-set limits and BNY’s Agentic Negotiation Protocol; humans approved the deal.
  • BNY plans to explore agent negotiations for repo deals, foreign-exchange payments, collateral terms and contracts.
  • Apple’s macOS 27.3 beta adds more granular app permissions, approval choices and a menu-bar indicator for app control; existing permissions remain in place.
  • Acemate added AI-generated tennis shot highlights with tracking and ball-speed and placement overlays to its iOS and Android app.
  • Ridge Security says RidgeGen uses checks, guardrails and reviewable evidence; automated remediation is planned for early next year.
  • Liquid AI’s d1.2-ultra scored 75 on the Artificial Analysis Decision Index v2.1, the highest score among decision models listed there.
  • OpenAI CEO Sam Altman warned that chip suppliers may rely on overlapping orders from buyers such as OpenAI and Anthropic, raising short-term supply-and-demand concerns.
  • Docusign’s ChatGPT tools prepare contract-renewal briefs using agreement details and CRM data such as Salesforce, and support legal review and agreement creation.
  • INFevo raised more than RMB 100 million in Pre-A funding for virtual-cell models and automated experiments; it is also developing tumor-organoid research with a Beijing laboratory.
  • Pepperdine’s two-day local-government AI workshop is scheduled for December 2–3, 2026; applications are due November 20.

BNY and Microsoft test AI agents in a Treasury trade

Bank of New York Mellon and Microsoft used AI agents to negotiate a US Treasury trade, which the companies called the first of its kind. The agents agreed on a price, and people approved the deal. Microsoft set limits for what its agent could accept, while BNY’s control layer let both sides confirm the terms. BNY calls the system the Agentic Negotiation Protocol and plans to test it in other negotiated market activities, including repo deals and foreign exchange payments. The agents still do not fully match human traders.

AI agents negotiate a first of its kind Treasury deal

Bank of New York Mellon and Microsoft used artificial intelligence agents to negotiate a US Treasury trade, then had people approve the transaction. Microsoft gave its agent boundaries for an acceptable deal, and BNY built a control layer so authorized agents could negotiate and confirm their understanding of the terms. BNY calls the system the Agentic Negotiation Protocol. The bank plans to keep testing it and explore other negotiated tasks, including repo deals, foreign exchange payments, collateral terms and contracts.

BNY and Microsoft complete an AI negotiated Treasury trade

Bank of New York Mellon and Microsoft used AI agents to negotiate a US Treasury trade, which they described as the first of its kind. Microsoft set boundaries for an acceptable deal, while BNY’s control layer allowed authorized agents to negotiate and confirm that both sides understood the terms. The agents agreed on a price, and humans approved the transaction. The system is called the Agentic Negotiation Protocol.

Apple tests new Mac privacy controls for AI agents

Apple is testing macOS 27.3, which changes how Mac apps get access to sensitive information and control over other apps. It splits the Device Control and Data Access setting into System & App Protection Override and Accessibility, and lets users limit access to selected folders or apps. Users can require approval for each request, and a menu bar indicator will show when one app controls another. Existing permissions will remain in place, and Apple says developers should update their apps to handle limited access. The changes follow reports about AI agents and other apps accessing private data.

macOS 27.3 adds tighter controls for AI agent access

Apple’s macOS 27.3, currently in beta 2, introduces new controls for apps that request sensitive data access or control over other apps. The update separates Device Control and Data Access into System & App Protection Override and Accessibility, and lets users choose specific folders or apps for supported permissions. Users can select Ask First or At Any Time for some app control requests, while a menu bar indicator shows when one app controls another. Previously granted permissions remain active, and developers must make sure their apps work with limited access. Apple says the changes aim to give users more visibility into the data and controls they grant to apps.

Acemate turns tennis training into AI highlight videos

Acemate added AI Shot Highlights to its tennis app for iOS and Android. The feature combines phone video with shot data, skeletal tracking, ball speed and placement overlays. Machine learning automatically highlights key moments to help players review their performance. The update also improves video quality, loading times and the user interface.

Pepperdine offers local leaders hands on AI training in December

The Davenport Institute for Public Engagement and Civic Leadership at Pepperdine University’s School of Public Policy will hold a two day workshop called Applied AI in Local Government: Practical Skills for Public Sector Leaders. The workshop will take place in Malibu on December 2 and 3, 2026. Applications are due Friday, November 20, 2026.

Ridge Security promotes its agentic AI security platform over open source tools

Ridge Security says its RidgeGen agentic AI security platform combines leading commercial large language models with the company’s own security knowledge, intelligence and testing data. It says the platform uses checks to verify findings, guardrails to keep testing within authorized limits, and logs and evidence that teams can review. Ridge Security also points to its research team, updates and technical support as advantages over open source tools. Automated remediation is planned for early next year, and the article says enterprises should consider security, maintenance, support, reliability, costs and governance when choosing an open source solution.

Liquid AI launches decision models with a top score on the AA index

Liquid AI released d1.2 and d1.2-ultra, models that accept text, images and audio. The release adds five new answer types. On the Artificial Analysis Decision Index v2.1, d1.2-ultra scored 75, the highest score among decision models listed there. Deployment options and specifications are available through the GPULeaderboard.com Chrome extension, according to the article.

AI agent rollouts depend on more than technology

After studying AI product teams, the author reviewed agent deployments in banks, utilities, manufacturers, retailers and other organizations outside the tech industry. The author found that nine practical rules described in an earlier August article mostly held up in these real world settings. The examples also point to common factors that help agents reach production beyond the technology itself. People, processes and data all matter when organizations design and deploy AI agents.

Sam Altman warns AI chip suppliers may count on the same buyers

At Semafor’s November 16 Silicon Valley & the World event, OpenAI CEO Sam Altman warned that many chip suppliers expect to sell large volumes to the same companies, including OpenAI and Anthropic. He said these buyers have limited demand and data center capacity, raising concerns that suppliers may be counting on the same future orders. Altman said data center demand remains solid and computing needs should grow over time. His concern is whether chip supply and demand will match in the short term, as AI stocks fell.

Docusign brings contract renewals and legal approvals into ChatGPT

Docusign has added tools that let sales teams prepare contract renewals and handle contracting tasks inside ChatGPT. The renewal brief combines signed agreement terms, deadlines and commitments with customer information from a connected CRM system such as Salesforce. Sellers can send requests to Legal for review through existing Docusign workflows or create agreements from customer data using company-approved templates. Docusign says its MCP server protects contract information with enterprise permissions, end-to-end encryption and customizable guardrails, including for actions such as creating agreements and collecting signatures.

INFevo raises over RMB 100 million to advance AI virtual-cell research

INFevo announced on November 16, 2026, that it raised more than RMB 100 million in Pre-A funding to expand its AI virtual-cell models and automated experiments. Xinglian Capital, Prosperity7 Ventures, Huagai Capital, Huakong Fund and DT Capital joined the round, while existing investors Shunwei Capital, HSG and Songhe Capital invested again. The company aims to use virtual cells to predict how cells respond to drug or genetic changes, then test predictions with automated lab experiments. Its Popper Project uses research agents and databases to develop and review testable hypotheses, and its Karl Workbench is in beta. INFevo is also working with a Beijing key laboratory led by Peking University First Hospital on liver, biliary and pancreatic tumor organoids, with the stated goal of connecting clinical questions, patient-derived models, drug screening, AI analysis and clinical validation.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

AI agents agentic negotiation financial services U.S. Treasury trading Microsoft BNY Mellon macOS privacy app permissions AI tennis highlights RidgeGen AI security Liquid AI multimodal AI AI chip demand data centers Docusign ChatGPT contract management virtual cell models automated experiments

Comments

Loading...