GitHub has rolled out a series of updates for its Copilot ecosystem, introducing new model options, enhanced security controls, and improved workflow features. The October 9, 2026 release brings Claude Haiku 5.5 to premium users and makes local sandboxing generally available across several Copilot environments.
The weekly update addresses practical friction points for developers, offering greater flexibility for enterprise users and deeper integration with local AI models.
What was announced
According to the official GitHub Changelog, the latest weekly release focuses on making the AI assistant easier to manage across different accounts and environments. The update spans the standalone Copilot application, the command-line interface (CLI), and the Visual Studio Code (VS Code) extension.
Highlighting the release is the integration of a new Anthropic model for paid tiers, alongside the general availability of local sandboxing to control exactly what AI agents can access on a developer’s machine.
Key details
- New model access: Claude Haiku 5.5 is now available for users on Copilot Pro, Pro+, Max, Business, and Enterprise plans.
- Local sandboxing: Now generally available at no extra cost in the Copilot CLI, the Copilot app, and VS Code sessions using Agent Host. This feature allows users to limit an agent’s access to local files, networks, and credentials.
- Dual-account support: The Copilot app now permits users to separate their GitHub accounts for licensing and repository access. For example, a developer can authenticate using an enterprise-provided Copilot license while working in repositories linked to a different personal or client account.
- Ollama integration in CLI: The GitHub Copilot CLI introduces a streamlined way to choose local models. By typing /model, users can discover and select supported models from a running local Ollama instance, operating them alongside configured cloud models.
- VS Code enhancements: In the VS Code 1.141 release, users can now view multiple agent sessions side-by-side in a grid, making it easier to compare results or monitor several tasks simultaneously. Additionally, a new “Open worktree cleanup” chat command helps developers see how much disk space inactive session worktrees are consuming and choose which ones to delete.
Why it matters
These updates provide developers with significantly more control over their AI coding environments. The ability to separate licensing and repository accounts in the Copilot app resolves a common enterprise hurdle, allowing developers to use company-funded AI tools on external or personal projects without complex workarounds.
Security is also a major focus. By making local sandboxing generally available, GitHub is establishing crucial guardrails for autonomous coding agents, ensuring they cannot indiscriminately access sensitive local credentials or system files while executing tasks.
Furthermore, the ability to discover local Ollama models directly in the Copilot CLI signals GitHub’s continued accommodation of offline and locally hosted AI workflows, giving developers the flexibility to blend local privacy with cloud-based capabilities.
Limitations and open questions
While the changelog confirms the addition of Claude Haiku 5.5 to most paid tiers, it does not specify if the model is subject to any distinct usage caps, rate limits, or regional restrictions compared to GitHub’s default models. Additionally, the announcement does not detail exactly which local models from Ollama are officially supported in the CLI discovery feature.
Sources
- GitHub Copilot weekly releases — October 5 — GitHub Changelog (official announcement), October 9, 2026
Related reading
- AI to Empower Developers, Says Embracer Group
- Intel and AMD’s Copilot Plus PCs to Launch Without Copilot AI Features
This article was prepared with AI assistance from the sources listed above and is subject to our editorial policy. Spot an error? Request a correction.