Microsoft Research Blog

Reducing Privacy leaks in AI: Two approaches to contextual integrity  

November 25, 2025
New research explores two ways to give AI agents stronger privacy safeguards grounded in contextual integrity. One adds lightweight, inference-time checks; the other builds contextual awareness directly into models through reasoning and RL.
Four white line icons on a blue-to-orange gradient background: a network node icon, a security shield with padlock icon, an information icon, a checklist icon

Subscribe to our newsletter

Sign up to get the latest

Recent Posts

  1. Four white line icons on a blue-to-orange gradient background: a network node icon, a security shield with padlock icon, an information icon, a checklist icon

    Reducing Privacy leaks in AI: Two approaches to contextual integrity  

    November 25, 2025

    New research explores two ways to give AI agents stronger privacy safeguards grounded in contextual integrity. One adds lightweight, inference-time checks; the other builds contextual awareness directly into models through reasoning and RL.

  2. Three white line icons on a blue-to-green gradient background: a computer monitor with a globe symbol on the left, a cursor arrow with click lines in the center, and a computer mouse outline on the right.

    Fara-7B: An Efficient Agentic Model for Computer Use 

    November 24, 2025

    Fara-7B is our first agentic small language model for computer use. This experimental model includes robust safety measures to aid responsible deployment. Despite its size, Fara-7B holds its own against larger, more resource-intensive agentic systems.

  3. Three white icons on a blue-to-purple gradient background: the first icon shows an image/photo; the second icon depicts a computer monitor with vertical bars; the third icon displays three connected circles with user silhouettes.

    MMCTAgent: Enabling multimodal reasoning over large video and image collections 

    MMCTAgent enables dynamic multimodal reasoning with iterative planning and reflection. Built on Microsoft’s AutoGen framework, it integrates language, vision, and temporal understanding for complex tasks like long video and image analysis.

  4. Three white icons on a blue-to-green gradient background: the first icon shows a circle with connected nodes, the second shows a circuit, and the third shows a flowchart

    BlueCodeAgent: A blue teaming agent enabled by automated red teaming for CodeGen AI 

    November 11, 2025

    BlueCodeAgent is an end-to-end blue-teaming framework built to boost code security using automated red-teaming processes, data, and safety rules to guide LLMs’ defensive decisions. Dynamic testing reduces false positives in vulnerability detection.

  5. Three white icons on a blue-to-purple gradient background: the first icon shows a node cluster, the second shows a person in front of a screen with another person, the third is a magnifying glass

    When industry knowledge meets PIKE-RAG: The innovation behind Signify’s customer service boost 

    November 6, 2025 |Industry Innovation Center

    A collaboration between Signify and Microsoft Research shows how PIKE-RAG improves enterprise knowledge systems, delivering a 12% increase in accuracy and faster, more reliable answers.

  6. Four white icons on a blue-to-purple gradient background: the first icon shows a node cluster, the second shows two persons, the third is a building, and the fourth is a location pin

    Magentic Marketplace: an open-source simulation environment for studying agentic markets 

    November 5, 2025

    AI agents are poised to transform digital marketplaces. To explore what can happen when AI agents interact and transact at scale, we built Magentic Marketplace, an open-source simulation environment for studying agentic market designs.

  7. white icons on blue and green gradient background

    RedCodeAgent: Automatic red-teaming agent against diverse code agents 

    November 4, 2025

    Code agents help streamline software development workflows, but may also introduce critical security risks. Learn how RedCodeAgent automates and improves “red-teaming” attack simulations to help uncover real-world threats that other methods overlook.

  8. Workflow icons showing tasks, thinking, and time, linked to a person symbol on a gradient background.

    Tell me when: Building agents that can wait, monitor, and act 

    October 21, 2025

    SentinelStep enables AI agents to handle monitoring tasks that run for hours or days, like watching for emails or tracking prices. It works by managing when agents should check and their context, avoiding wasted resources and missed updates.

  9. Paraphrase Project Protiens

    When AI Meets Biology: Promise, Risk, and Responsibility 

    October 6, 2025 |Eric Horvitz

    Microsoft researchers reveal a confidential research effort that explored how open-source AI tools could be used to bypass biosecurity checks—and helped create fixes now influencing global standards.

  10. Icons representing individual and group connections to a central computer monitor with a globe, symbolizing online connectivity, set against a gradient background transitioning from blue to pink.

    Using AI to assist in rare disease diagnosis 

    September 22, 2025 |Mandi Hall andAshley Conard

    New research from Microsoft, Drexel, and the Broad explores how generative AI could support genetic professionals in rare disease diagnosis.

  11. Three white icons on a gradient background transitioning from blue to purple to pink. From left to right: a globe with a magnifying glass representing internet search, a central circle connected to smaller circles symbolizing network connectivity, and a checklist with two checkmarks and one empty box indicating task management.

    Tool-space interference in the MCP era: Designing for agent compatibility at scale 

    As agentic AI ushers in a new era marked by tool expansion, systems are converging, and complexity is rising. Microsoft Research explores the Model Context Protocol (MCP) as a new standard for agent collaboration across fragmented tool ecosystems.

  12. Three white icons on a gradient background transitioning from blue to green. From left to right: network node icon, lightbulb-shaped icon with a path tool icon in the center; a monitor icon showing a web browser icon

    RenderFormer: How neural networks are reshaping 3D rendering 

    September 10, 2025 |Yue Dong

    RenderFormer, from Microsoft Research, is the first model to show that a neural network can learn a complete graphics rendering pipeline. It’s designed to support full-featured 3D rendering using only machine learning—no traditional graphics computation required.

Explore More