
Microsoft 365 E7 Guardian Agents Apply Dynamic Permissions to Mitigate AI Insider Threats
New tools treat autonomous AI agents as potential insider threats, using continuous risk assessments to manage permissions and access controls.
Everything Vector Bulletin has reported about Microsoft: 18 stories, newest first. Part of our Tech News coverage.

New tools treat autonomous AI agents as potential insider threats, using continuous risk assessments to manage permissions and access controls.

Microsoft introduces adaptive security controls for AI agents within Project Perception to contain risks and enforce policy boundaries.

Microsoft proposes a new security architecture to supervise autonomous AI agents operating within enterprise networks.

Microsoft has released a set of ten new desktop backgrounds for Windows Insiders, celebrating twelve years of the program with designs inspired by the Windows 11 Bloom aesthetic.

Microsoft unveils new hardware alongside NVIDIA chips, but reviewers note the operating system may limit the device's overall value proposition.

Multiple Windows 11 cumulative updates are failing to install, displaying a specific error code that has persisted for several months.

A new joint whitepaper from Microsoft and NVIDIA provides enterprises with a shared taxonomy and controls to manage risks in open-weight AI deployments.

Five major US technology firms commit to spending between $660 billion and $690 billion on AI hardware and data centres next year.

Documents reveal Nvidia and Apple each committed $25 million, the highest amounts, to fund the construction of Donald Trump’s White House ballroom.

Microsoft publishes open-source Integrated HSM technology to enhance hardware security and cryptographic key protection through industry collaboration.

A high-severity memory corruption bug in Microsoft Excel allows remote attackers to run arbitrary code on affected systems without prior authentication.

Microsoft has released patches for a high-severity denial-of-service vulnerability in .NET that allows remote attackers to crash services via uncontrolled recursion.

Microsoft has released updates for System.Security.Cryptography.Xml to fix a high-severity flaw that allows attackers to deny service via unbounded resource use.

Microsoft has launched a small AI model designed to select from fixed answer options, built upon Alibaba’s open-weight Qwen3.5-9B foundation.

A critical flaw in Microsoft QUIC allows unauthenticated remote code execution via crafted IPv6 packets, affecting .NET 8 through 11 on Windows.

An unauthenticated remote code execution vulnerability in Windows Message Queuing affects multiple Windows 10 and 11 builds with a CVSS score of 9.8.

The new model returns probabilities for fixed answer sets, offering lower latency and cost for structured decision workflows compared to general-purpose LLMs.

Microsoft assigns a high severity score to an information disclosure vulnerability in its cloud API management service that permits unauthorised network access.