30+ sources. Zero spin.
Cross-referenced, unbiased news. Both sides of every story.
Microsoft Build Day Two: Scout AI Gets a Name and a VP, MXC Locks Down Agent Security, and a New Reasoning Model Enters the Race

Scout Gets a Face — and a Reality Check
Microsoft's always-on Teams AI agent now has an official executive owner: Omar Shahine, newly appointed corporate vice president of Microsoft Scout, per Wired.
Shahine's pitch is simple. "Your company essentially hires your assistant," he told Wired. The agent monitors your email, calendar, and Teams messages, blocks family dinner from being invaded by all-hands meetings, and tracks every promise you've made or received.
Shahine's own Scout agent — which he nicknamed Sebastian — sent an email that was "just one big run-on sentence, no formatting." The VP of the product is debugging it in real time. Enterprises should note: Scout is rolling out to a small group of customers first, and a desktop app is live today only for users with "frontier" feature access AND an active GitHub Copilot subscription. Two paywalls deep.
MXC: The Security Story Everyone Is Underreporting
Microsoft's Execution Containers, or MXC, is one of the most consequential announcements at Build this year.
According to VentureBeat, MXC is an OS-level, policy-driven execution sandbox built directly into Windows — and the Windows Subsystem for Linux — that enforces exactly what an AI agent can and cannot access at the kernel level. Every agent gets a bound identity, either a local ID or a Microsoft Entra cloud identity, so every action is attributed and auditable.
OpenAI and Nvidia are already on board as launch partners, according to VentureBeat.
The enterprise AI deployment problem isn't "can the agent do the task." It's "what happens when the agent does something it wasn't supposed to do." MXC is Microsoft's answer — not by hobbling agents, but by building the walls before you let them run. No other major platform has shipped this at the OS level.
MAI-Thinking-1: Microsoft's First Reasoning Model Is Here, and It's Late
Microsoft AI CEO Mustafa Suleiman announced seven new AI models at Build. The headline among them is MAI-Thinking-1, the company's first-ever reasoning model.
The specs, per ZDNET: 35 billion parameters, trained on "enterprise-grade, clean and commercially licensed data" — a deliberate shot in an era of active AI copyright lawsuits. Microsoft says it beat Anthropic's Sonnet 4.61 in a blind evaluation and matches Anthropic Opus 4.6 on the SWE Bench Pro coding benchmark.
But ZDNET's Radhika Rajkumar noted: "Most labs have created multiple advanced reasoning models at this point, so it's unclear yet where Microsoft's approach stands relative to those."
MAI-Thinking-1 is in private preview on Microsoft Foundry. It's NOT publicly available yet.
Also shipping: MAI-Code-1, described as "ultra-efficient" and tuned for GitHub, coming to Copilot and VS Code now. MAI-Image-2.5 is Microsoft's first text-to-image model — live in PowerPoint and Foundry — and reportedly hit third place on the LM Arena Leaderboard the moment Suleiman announced it on stage, per ZDNET.
Microsoft IQ and the Data Silo Problem
Every new AI agent your company deploys starts from zero organizational context. VentureBeat has been covering this as a central enterprise AI challenge.
Microsoft's answer is Microsoft IQ — an expansion of Fabric IQ into a four-layer unified context system: Work IQ (how your organization operates day to day), Foundry IQ (institutional knowledge and rules), Fabric IQ (live operational data), and Web IQ (real-time global signals). Any agent connects to all four in a single integration step.
VentureBeat's VB Pulse Q1 2026 data shows hybrid retrieval intent among 100-plus employee organizations tripled from 10.3% in January to 33.3% in March. Enterprises are deploying more AI agents while trying to prevent them from hallucinating different answers to the same question depending on which agent asks.
Paired with this is Rayfin, a new open-source SDK that deploys agent-built apps directly into Fabric as a governed backend — so new applications route data into the existing platform instead of spawning yet another silo.
The Hardware Angle: Surface RTX Spark Dev Box and the Death of the Per-Token Bill
Microsoft debuted the Surface RTX Spark Dev Box — a compact desktop powered by Nvidia's Blackwell RTX Spark chip with 128GB of unified memory, capable of running models exceeding 120 billion parameters with ZERO cloud API calls, according to VentureBeat.
Executive VP Pavan Davuluri explained the engineering logic: at 100,000 tokens of context, the key-value cache alone eats 40-50GB of memory. That's why 128GB unified memory isn't a vanity spec. It's a requirement.
No pricing disclosed. Available later this year in the US, sold exclusively through Microsoft.com.
Meanwhile, Nvidia's RTX Spark chip is coming to laptops from HP, Lenovo, Acer, Asus, Dell, MSI, and Microsoft's own Surface Laptop Ultra — a 15-inch flagship ZDNET describes as the most powerful Surface laptop ever built. Expect all of them to clear $2,000, per ZDNET's Kyle Kucharski.
MDASH Exits Preview — 100+ Threat-Hunting Agents, Now Fully Integrated
Microsoft's agentic security scanning system, MDASH, is out of preview and is now folded into a full enterprise security control plane connecting Defender, GitHub Code Security, Agent 365, and Purview, according to ZDNET.
Microsoft's chief security architect Aleš Holeček told ZDNET: "AI vulnerability discovery has crossed from research curiosity into production-grade defense at enterprise scale."
The core value proposition: stop drowning security teams in thousands of low-priority scanner alerts. MDASH does triage — it surfaces only real, exploitable vulnerabilities. Security alert fatigue is a known, documented cause of breaches.
Microsoft's Build Agenda
Microsoft dropped substantial announcements at Build across multiple fronts: Scout and new models drew immediate attention, while MXC's OS-level agent security, Microsoft IQ's context unification, and MDASH's triage architecture represent the underlying infrastructure for agent deployment at scale. For enterprises deploying AI agents in 2026, these foundational elements determine whether the technology becomes manageable or ungovernable.