Microsoft announced agentic features for Copilot in Outlook, expanding from single-task assistance to continuous, multi-step automation of email and calendar work. These features let Copilot act independently on your behalf — prioritizing messages, drafting follow-ups, responding to meeting invites, and resolving scheduling conflicts. Access is currently limited to Microsoft's Frontier early-access program and requires a Microsoft 365 Copilot license. This article explains what the new features do, what your infrastructure must look like, and how you enable access as an administrator.

Source

At FabCon and SQLCon 2026 in Atlanta, Microsoft announced additions to Microsoft Fabric, its integrated data and analytics platform. The highlights include the Database Hub—a unified management console for major Microsoft database services—Fabric IQ, a new semantic layer that gives AI tools a structured understanding of business data, and a series of updates to OneLake, the shared data storage layer that underpins Fabric. This article provides simple introductions to OneLake, Fabric, Database Hub, and Fabric IQ, and outlines other new features announced in Atlanta.

Source

Microsoft released new security and governance controls for Microsoft 365 Copilot, introducing Data Loss Prevention (DLP) policies that inspect prompts before Copilot processes them, protecting web searches from leaking sensitive data, and enabling bulk remediation of overshared SharePoint files. The Copilot Dashboard gained expanded access, user satisfaction tracking, and CSV export. This article explains what each feature does technically, how to configure the relevant policies, and where the current limitations are.

Source

Microsoft's Conditional Access Optimization Agent uses AI to continuously scan your Microsoft Entra ID environment for policy gaps and recommend remediations. It requires a Microsoft Entra ID P1 license and a Microsoft Security Copilot subscription. Several new features are in public preview, including context-aware recommendations, deep gap analysis across all policies, and phased policy rollouts. This article explains what Conditional Access policies are, how the agent works, what you need to run it, and where its limitations lie.

Source

GitHub Copilot CLI—GitHub's AI assistant for the terminal—introduced an experimental feature called Rubber Duck that pairs your primary AI model with an independent reviewer from a completely different AI vendor. The goal is to catch early planning mistakes before they compound into harder-to-fix downstream errors, a known weakness of single-model AI agents. In performance testing on real-world coding problems, combining Claude Sonnet with Rubber Duck closed 74.7% of the performance gap between Sonnet and the more capable Opus tier.

Source

The Model Context Protocol (MCP) is an open-source standard for connecting AI agents to external resources such as database systems. MCP extensions are modular, optional additions to the Model Context Protocol that enable specialized capabilities without modifying the core specification. These extensions support features like interactive user interfaces, custom authentication flows, and industry-specific functionality while maintaining compatibility with the base protocol.

Source

Seven protocols have emerged to standardize communication between AI agents, models, and external systems: Anthropic Model Context Protocol (MCP), Google Agent2Agent (A2A), Google Agent Gateway Protocol (AGP), Cisco AGNTCY, IBM Agent Communication Protocol (ACP), and Zed Agent Client Protocol (AGP). This article provides an overview of these AI protocols and compares their use cases.

Source

Microsoft Agent 365 is the control plane for AI agents in the enterprise: a set of capabilities built on Microsoft Entra, Purview, Defender XDR, and the Microsoft 365 Admin Center that gives IT and security teams a single location to observe, govern, manage, and secure AI agents — regardless of whether they were built with Microsoft tooling, open-source frameworks, or third-party platforms.

Source

Microsoft announced Wave 3 of Microsoft 365 Copilot, introducing Copilot Cowork as a new AI-driven work mode that executes multi-step tasks across Office applications. However, Copilot Cowork is a double misnomer because the tool qualifies as neither a copilot nor a coworker.

Source

Microsoft has completed the global rollout of AI workflow templates in the Teams Workflows app, powered by Microsoft 365 Copilot, as of mid-February 2026 (Roadmap ID 500379). The feature introduces scheduled prompts — recurring, AI-driven automations that run without manual intervention. You can use pre-built templates to automate tasks such as weekly project summaries or data monitoring across the Microsoft 365 ecosystem. The feature is restricted to users with a Microsoft 365 Copilot license and is currently available only in Teams for Web and Teams for Mac.

Source

Autonomous AI agents can perform complex tasks without human intervention, but face a critical challenge: how can they trust agents from other organizations? ERC-8004, developed by teams from the Ethereum Foundation, Google, Coinbase, and MetaMask, addresses this by providing a decentralized identity, reputation, and verification system for AI agents. Launched on Ethereum mainnet on January 29, 2026, the standard enables agents to discover reliable service providers and verify past performance. Over 20,000 AI agents were deployed across multiple blockchains within the first two weeks.

Source

ChatGPT Lockdown Mode is an advanced security setting designed to protect organizations against prompt-injection-based data exfiltration attacks. Introduced by OpenAI in February 2026, you can enable this optional feature through workspace settings if you manage enterprise ChatGPT accounts.

Source

WebMCP is a proposed web standard that enables websites to expose structured, callable tools directly to AI agents through a new browser API. Developed jointly by engineers at Google and Microsoft and incubated through the W3C's Web Machine Learning Community Group, WebMCP allows you to wrap your existing client-side JavaScript logic into agent-readable tools without re-architecting your web pages. Instead of AI agents relying on screen scraping or DOM manipulation to interact with websites, WebMCP provides a standardized way for agents to discover and invoke specific functions with defined schemas, ensuring more reliable and precise interactions.

Source

Anthropic announced on February 12, 2026, that Claude Enterprise is now available for direct self-serve purchase on its website, eliminating the need for sales conversations. Organizations can set up their workspace, configure single sign-on, and begin inviting team members within minutes.

Source

Visual Studio Code version 1.109 introduces agent orchestrations — a feature that transforms the IDE into a multi-agent development platform. You can now create custom agents that collaborate, delegate tasks to specialized subagents, and execute complex workflows autonomously. This article explains the concept and demonstrates agent orchestration with a simple example. I'll also share my opinion on Microsoft's concept of agent orchestrations. It's not particularly favorable.

Source

OpenClaw, previously known as Clawdbot and briefly as Moltbot, is a free, open‑source personal AI agent that runs locally on your computer, enabling you to automate any task a human can perform on a computer. It gained rapid popularity, accumulating 100,000 GitHub stars in record time, along with tens of thousands of forks as it went viral. OpenClaw skills pose severe security risks because they are unvetted modules that may contain hidden, malicious instructions for the agent's model. These harmful instructions could result in API key leaks, data theft, credential compromise, and system breaches due to malicious code execution. This guide covers OpenClaw skill structure, known security incidents, and step-by-step instructions for scanning agent skills with the Cisco AI Skill Scanner. The Cisco AI Skill Scanner provides comprehensive security analysis of OpenClaw agent skills before installation, detecting malware, data exfiltration, and prompt-injection vulnerabilities.

Source

Microsoft released version 1.2.0 of the Microsoft Entra PowerShell module, introducing production-ready support for Agent Identity Blueprints, enhanced application configuration parameters, and modernized invitation APIs. This update consolidates Agent Identity functionality into the main module and delivers new cmdlets for automated identity management across Microsoft Entra ID environments.

Source