On April 3, 2026, Anthropic rolled out Microsoft 365 connectors to all Claude subscription plans. Meanwhile, GitHub published two security improvements for the Copilot cloud agent: commits are now signed and verified, and the firewall can be managed at the organization level. Gemini CLI released version 0.36.0 with native sandboxing and support for git worktrees, while the Manus agent became accessible via LINE.
Microsoft 365 connected to all Claude plans
April 3 โ Microsoft 365 connectors (Outlook, OneDrive, SharePoint) are now available on all Claude plans: Free, Pro, Max, Team and Enterprise.
Previously reserved for premium subscriptions, these connectors let you bring emails, Word documents, Excel sheets and SharePoint files directly into a Claude conversation without manual copy-paste. Activation is done from the customization settings.
Microsoft 365 connectors are now available on every Claude plan. Connect Outlook, OneDrive, and SharePoint to bring your email, docs, and files into the conversation.
Microsoft 365 connectors are now available on every Claude plan. Connect Outlook, OneDrive and SharePoint to bring your email, docs, and files into the conversation. โ @claudeai on X
Copilot cloud agent: signed commits and organization-level firewall
April 3 โ GitHub published two security improvements for the Copilot cloud agent in a single day.
Signed commits (Verified badge)
The Copilot cloud agent now signs every commit it creates. Commits appear with the โVerifiedโ badge on GitHub, which guarantees their authenticity.
The main practical impact is that the agent can now work in repositories that enforce the branch protection rule โRequire signed commitsโ. Previously, this rule completely blocked the agent.
๐ Changelog โ signed commits
Organization-level firewall management
The Copilot cloud agent firewall โ which protects against prompt injection and data exfiltration โ can now be configured at the organization level, not just repo-by-repo.
Organization administrators can:
| Setting | Description |
|---|---|
| Global enablement | Enable or disable the firewall for all repositories |
| Recommended allowlist | Enable or disable the allowlist recommended by GitHub |
| Custom org allowlist | Add entries that cover all repositories (e.g., internal package registry) |
| Delegation to repo admins | Allow or deny repo admins the ability to add their own entries |
By default, each repository keeps its current settings โ migration is not forced.
๐ Changelog โ organization firewall
Gemini CLI v0.36.0: native sandboxing and Git worktrees
April 1 โ Google released Gemini CLI version 0.36.0 with several security and productivity improvements.
Native sandboxing โ The multi-registry architecture for subagents is accompanied by native sandboxing on the two main platforms: macOS Seatbelt and Windows. These mechanisms isolate subagent execution to limit the risk of unauthorized filesystem access.
Git worktree support โ Gemini CLI natively supports git worktrees, allowing multiple isolated working sessions in parallel on the same repository. Useful for working on multiple branches simultaneously.
Refreshed Composer interface โ The Composer interface has been redesigned for better readability and more reliable terminal interaction.
JIT context injection for subagents โ Subagents benefit from Just-In-Time context injection, with a more resilient tool-rejection system that provides contextual feedback on errors.
| Feature | Description |
|---|---|
| Multi-Registry Architecture | Multi-registry architecture for subagents |
| macOS Seatbelt | Native macOS sandboxing |
| Windows Sandboxing | Native Windows sandboxing |
| Git Worktree Support | Parallel isolated sessions on the same repo |
| JIT Context Injection | Context injected on demand for subagents |
| Refreshed Composer UX | Refreshed Composer interface |
๐ Gemini CLI Changelog
Manus available on LINE
April 3 โ Manus (general AI agent) is accessible from the LINE messaging app, popular notably in Japan.
Activation is done in three steps: go to the โAgentsโ tab in Manus, select โLINEโ and scan the QR code with your smartphone, then send a message once the account is linked. The announcement was made in coordination with the Japanese account @ManusAI_JP. Manus was already accessible via Telegram.
Claude Code v2.1.91 โ MCP extended and plugins bin/
April 2โ3 โ Claude Code moved to version 2.1.91 with additions on the MCP side and plugins.
The most useful change concerns MCP tool results: it is possible to force retention of a large result up to 500,000 characters by annotating the response with _meta["anthropic/maxResultSizeChars"]. This solves truncation cases for database schemas or large API responses.
Plugins can now bundle their own executables in a bin/ folder, callable directly from the Bash tool. On the fixes side, --resume no longer loses history when silent asynchronous write failures occur.
| Feature | Description |
|---|---|
| MCP result size override | Up to 500K characters via _meta["anthropic/maxResultSizeChars"] |
| Plugins bin/ | Bundled executables callable from the Bash tool |
| Multi-line deep links | %0A accepted in claude-cli://open?q= |
disableSkillShellExecution | Disables inline shell execution in skills and plugins |
GitHub Copilot โ deprecation of GPT-5.1 Codex and CLI metrics
April 3 โ Two practical updates for teams using Copilot.
Deprecation of GPT-5.1 Codex (effective April 1) โ The GPT-5.1-Codex, GPT-5.1-Codex-Mini and GPT-5.1-Codex-Max variants are replaced by GPT-5.3-Codex in Copilot Chat, inline suggestions, and agent modes. Enterprise administrators must enable GPT-5.3-Codex in model policies; no action is required to remove the old models.
๐ Changelog โ GPT-5.1 Codex deprecation
Per-user Copilot CLI metrics (April 2) โ Organization reports now include Copilot CLI activity per individual user for 1-day and 28-day reports. The new metrics cover: sessions, requests, tokens used, and last known CLI version.
๐ Changelog โ CLI metrics
What this means
Opening Microsoft 365 connectors to all Claude plans is the most directly useful decision of the day: it removes the subscription barrier to connecting your work environment to Claude, which until now limited usage to premium subscribers.
On GitHubโs side, the two security improvements for the Copilot cloud agent head in the same direction: making the agent operable in enterprise environments with strict security policies. Signed commits and organization-level firewall settings address concrete blockers enterprise teams faced.
Gemini CLI 0.36.0 confirms the chosen direction: native sandboxing on the two main platforms and support for Git worktrees for parallel workflows. These additions position Gemini CLI as a serious tool for teams working on multiple branches simultaneously.
Sources
- @claudeai on X โ M365 connectors
- GitHub Changelog โ Copilot signed commits
- GitHub Changelog โ Copilot organization firewall
- Gemini CLI Changelog
- @ManusAI on X โ Manus on LINE
- CHANGELOG Claude Code v2.1.91
- GitHub Changelog โ GPT-5.1 Codex deprecation
- GitHub Changelog โ Copilot CLI metrics
This document was translated from the fr version into the en language using the gpt-5-mini model. For more information on the translation process, consult https://gitlab.com/jls42/ai-powered-markdown-translator