Microsoft AI Updates: October 11, 2026

1. GitHub Copilot Makes Local Agent Sandboxing Generally Available

GitHub Copilot. GitHub’s weekly Copilot release made local sandboxing generally available in Copilot CLI, the Copilot app, and VS Code sessions using Agent Host, letting users limit an agent’s access to files, networks, and credentials at no extra cost. Copilot CLI’s /model command can now discover models from a running local Ollama instance alongside configured and cloud models, and the Copilot app now supports using one GitHub account for the Copilot license and a separate account for repository access. The same release added Claude Haiku 5.5 for Copilot Pro, Pro+, Max, Business, and Enterprise users, and VS Code 1.141 adds a grid view for running agent sessions side by side plus a worktree cleanup command for inactive sessions. Source

2. Copilot for JetBrains Adds Enterprise-Managed Default Models and MCP Startup Controls

GitHub Copilot. The GitHub Copilot plugin for JetBrains IDEs now lets enterprise administrators use managed settings to set any available Copilot agent model as the default for new conversations, while users keep an explicit model picker. Diagnostic intention menus gain a Fix action that opens inline chat and uses agent mode when available, and a new setting disables automatic MCP server startup for Copilot and Claude. The release also drops support for JetBrains IDE 2025.1, so the plugin now requires version 2025.2 or later. Source

3. Microsoft Launches Decision-1, a Single-Pass Decision Model on Foundry

Microsoft. Microsoft released Decision-1, a model post-trained from Qwen3.5-9B that returns a calibrated probability for each option in a fixed set in one forward pass, supporting yes/no, multiple-choice, rating, and rubric-based grading for classification, routing, and agent control. Microsoft reports the highest accuracy in its own 36-benchmark comparison covering nearly 150,000 held-out questions, decision flips on 1.3 percent of perturbed inputs, and P50 latency 2.5 times lower than the runner-up, H2O-Lightning-4B v1.1. It is available on Microsoft Foundry and OpenRouter at $0.042 per million input tokens with free output tokens, and Microsoft plans to rebase it on its MAI and OpenAI models. Source

4. Satya Nadella Calls for an “Emergency Brake” Architecture for AI Models

Microsoft. In a Saturday post on X, Microsoft CEO Satya Nadella called for reassessing the “trust architecture” of AI by separating the model from the orchestration harness that manages its work and moving safeguards outside the model. He proposed tamper-proof, human-readable logs of every significant model action, authority for an authorized person to pause or stop a model mid-task, and treating models as potentially compromised and contained from the start. The post followed a week of reports about AI labs losing control of agents during evaluations. Source