Browse Artificial Intelligence News (430)

Natalie Guevara summarizes GitHub’s April 2026 availability incidents, including outages and degradations affecting code search, audit logs, Copilot services, Pages, Codespaces, Actions, and other platform features, with root causes and concrete follow-up actions to improve detection, resilience, and recovery.

Imran Siddique and Shawn Henry explain how Microsoft Agent Framework and the Agent Governance Toolkit (AGT) fit together to run AI agents safely in production, with deterministic runtime policy checks, budget enforcement, and end-to-end auditability across local and cross-boundary (A2A) agent interactions.

Allison announces a technical preview of the GitHub Copilot app, a GitHub-native desktop experience for running focused, isolated agentic coding sessions from issues and pull requests, then validating changes and landing them through normal PR review and checks.

Alyssa Ofstein and Elliot H Omiya explain how defense in depth needs to adapt for autonomous AI agents, focusing on application-layer controls that bound what agents can do, how they get permissions, when humans must approve actions, and how identity makes agent behavior auditable.

Jim Harrer announces the VSLive! Microsoft AI Hackathon 2026 at Microsoft HQ in Redmond, a hands-on evening build event designed to help teams ship real prototypes using Azure OpenAI, Azure AI Foundry, GitHub Copilot, and agent-based patterns, with judging criteria that emphasize architecture, security, and practical value.

stclarke summarizes SAP Sapphire 2026 announcements focused on running SAP workloads on Azure and moving enterprise AI from pilots to production, including Azure OpenAI + Copilot Studio scenarios, Microsoft Fabric connectivity to SAP data, sovereign cloud options for regulated industries, and Sentinel-based monitoring for SAP landscapes.

Microsoft Defender Security Research Team and Yossi Weizman break down real-world “exploitable misconfigurations” in cloud-native AI apps—especially Kubernetes deployments where exposed services and weak auth can lead to RCE, credential theft, and data leaks—and show what to harden and what Defender for Cloud can detect.

Allison announces that GitHub Copilot’s cloud agent now supports Auto model selection, letting Copilot pick the best available model based on system health and performance, with a discounted model multiplier and no weekly rate-limit impact.

The Visual Studio Code Team shares the 1.121 (Insiders) release notes, covering Copilot Chat and agent terminal improvements, model picker updates, terminal output compression for common dev tools, SSH authentication enhancements, and a newer bundled ConPTY on Windows.

Allison announces updates to GitHub Copilot for JetBrains IDEs, including a public preview of the Copilot CLI agent inside the IDE, a unified sessions view for tracking agent runs, and new agent-mode capabilities like an Ask question tool and global custom agent configuration.

Shreyas Canchi Radhakrishna announces a preview update to Microsoft Fabric Data Agent that expands how it can query Eventhouse KQL databases by understanding user-defined functions, materialized views, and shortcut tables, improving consistency, performance, and access to external data sources.

Allison announces a public preview REST API that lets Copilot Business and Copilot Enterprise customers start and track Copilot cloud agent tasks programmatically, enabling custom automations that fan out work across repos and open pull requests with validated changes.

Amir Jafari announces preview support for service principal (SPN) authentication for Microsoft Fabric data agents, explaining why app identities matter for production deployments and outlining two common patterns: calling a data agent directly from a custom app, or routing through a Microsoft Foundry agent.

Simona Liao introduces Agent Skills in Visual Studio, a way to package reusable, task-specific instructions so Copilot agents can follow your team’s workflows and standards. The post shows how to create and manage skills in the IDE or from your repo, and when to use skills versus custom instructions.

Satya Nadella shares an update on Microsoft’s multi-model agentic security system, which uses 100+ specialized agents across frontier and custom models to find exploitable bugs, topped the CyberGym benchmark, and helped identify and fix 16 vulnerabilities ahead of Patch Tuesday, with a private preview now available.

Taesoo Kim announces MDASH, Microsoft Security’s multi-model agentic scanning harness, and explains how it uses specialized AI agents to find, validate, and prove vulnerabilities end-to-end. The post shares benchmark results, details 16 Patch Tuesday CVEs found in Windows networking/auth components, and includes two technical deep dives.

Taesoo Kim introduces MDASH, Microsoft’s multi-model agentic scanning harness, and explains how it’s being used to find and validate real Windows vulnerabilities end-to-end. The post breaks down the pipeline stages (prepare/scan/validate/dedup/prove), shares benchmark results, and details 16 Patch Tuesday CVEs plus two technical deep dives.

stclarke summarizes Microsoft and Red Hat’s Red Hat Summit 2026 updates for Azure Red Hat OpenShift, focusing on running modern apps and production AI with enterprise governance. It highlights OpenShift Virtualization for VM-to-Kubernetes migration, identity and confidential computing features, GPU-backed AI workloads, and expanded regional availability.

Allison announces improvements to GitHub Copilot code review comments in pull requests, aimed at making feedback easier to scan and act on with severity labels and grouped suggestions to reduce repetitive noise.

Allison announces that April usage reports are now available so GitHub Copilot admins and individual users can estimate how activity maps to AI credits ahead of the June 1 move to usage-based billing, including known gaps and data-quality issues in the report.

Natalie Guevara announces updates to GitHub Copilot’s individual plans ahead of the June 1, 2026 move to usage-based billing, including new “flex allotments” for Pro and Pro+ and a new Max plan for higher-volume usage.

samkemp announces Foundry Local 1.1.0, adding on-device live speech transcription, text embeddings for semantic search/RAG, and an Open Responses API client for streaming, tool calling, and vision. The post also covers WebGPU as an optional plugin, smaller JavaScript packages, and broader .NET compatibility for the C# SDK.

Lee Reilly explains how he used GitHub Copilot CLI—especially /delegate—to build “GitHub Dungeons”, a GitHub CLI extension that turns any repository into a terminal roguelike. The post covers the core idea (seeded by commit SHA), how Copilot’s agent workflow fit into iteration, and the BSP approach used for dungeon generation.

Nick Brady’s April 2026 digest covers Microsoft Foundry updates for model access, local inference, agent observability, and SDK changes across Python, JavaScript/TypeScript, .NET, and Java, with concrete guidance on quota tiers, tracing via OpenTelemetry, and monitoring/evaluation features for production agents.

stclarke summarizes the April 2026 Copilot Studio updates, focusing on scaling AI agents with stronger governance, clearer analytics visibility, and more capable workflows. It also covers new integration options like apps-in-agents, MCP-enabled tools (preview), evaluation automation APIs, and multi-agent collaboration features.

Kedasha Kerr explains what open source is and walks beginners through finding beginner-friendly repositories on GitHub, evaluating whether a project is well maintained, and making a first contribution using a fork-and-pull-request workflow (with an example prompt for GitHub Copilot Chat).

Shireesh Thota summarizes the main architecture trends from Cosmos DB Conf 2026, focusing on how teams are building AI-native apps on Azure Cosmos DB with flexible data models, serverless scale, and first-class semantic/vector search, plus practical patterns for agent memory, cost visibility, and multi-user security.

Allison announces that Grok Code Fast 1 will be deprecated across GitHub Copilot experiences on May 15, 2026, and outlines what Copilot Enterprise admins need to do to ensure alternative models are available for users.

Allison announces an update to the GitHub Copilot usage metrics API that adds a breakdown of Copilot code review suggestions by comment type, helping enterprises and organizations understand what kinds of review feedback Copilot generates and how often developers apply it.

Allison announces new “Agents” secrets and variables for Copilot cloud agent, enabling org-level configuration and finer repository access control so teams can roll out shared settings (like registry tokens or MCP server config) across many repos without duplicating Actions environment setup.

Jacob Alber walks through the Microsoft Agent Framework “Handoff” orchestration pattern for multi-agent workflows, showing how to model agent-to-agent routing as a bounded graph with shared conversation context, guardrails, and natural termination. The post compares .NET and Python authoring APIs and highlights when to choose Handoff vs Sequential or explicit conditional branching.

Allison announces that GPT-4.1 will be deprecated across GitHub Copilot experiences on 2026-06-01, and points teams to GPT-5.5 as the suggested replacement. The post highlights where admins may need to adjust Copilot Enterprise model policies so the alternative model is available in Copilot Chat.

Landon Cox explains how GitHub instrumented GitHub Agentic Workflows to track LLM token usage in CI, then used automated “auditor” and “optimizer” workflows to reduce costs. The post covers token-usage logging, MCP tool pruning, replacing MCP calls with GitHub CLI steps, and an “Effective Tokens” metric to compare savings across models.

Allison announces that Claude Sonnet 4 was deprecated across GitHub Copilot on May 6, 2026, and points teams to Claude Sonnet 4.6 as the replacement, with notes for Copilot Enterprise admins on enabling alternative models via model policies.

Kristen Womack introduces an Azure Developer CLI (azd) template from Curity and Microsoft that deploys an AI agent app to Azure with least-privilege authorization. It focuses on using short-lived OAuth 2.0 tokens (JWTs) and token exchange so APIs can enforce data boundaries even when agent behavior is nondeterministic.

Andrea Griffiths shares a practical checklist for reviewing agent-generated pull requests, focusing on where AI-written changes tend to hide risk: weakened CI, duplicated utilities, subtle logic bugs that still pass tests, and unsafe LLM-powered workflows that can turn untrusted input into executed commands.

The Visual Studio Code Team shares the 1.120 (Insiders) updates, focusing on chat/agent workflow improvements, model context controls, safer handling of terminal password prompts, and new extension APIs like custom diff editors with unified diff rendering.

Daniel Roth explains how Microsoft Copilot Studio upgraded its in-browser .NET WebAssembly engine from .NET 8 to .NET 10, what changed in deployment and packaging, and the performance gains the team measured—especially for larger, more complex agents running AOT-compiled code.

Juan Montes reports on how Porsche Cup Brasil built an AI-assisted crash analysis and telemetry workflow on Microsoft platforms, cutting damage assessment time and improving race operations with human-in-the-loop validation.

Allison announces expanded model support for Rubber Duck in GitHub Copilot CLI, enabling cross-model “second opinion” reviews by pairing GPT and Claude models depending on which model is orchestrating the session.