The GitHub Copilot app can now sandbox local sessions, but the setting is off by default
Since September 23, 2026, the GitHub Copilot app can limit what a local agent session reaches on your machine: folders, network access, and Git and GitHub CLI credentials. The feature is in public preview and has to be switched on for each project. A day earlier the app gained OpenTelemetry export for administrators, and Copilot code review settings reached every plan.
Source
Local sandboxing in the GitHub Copilot appGitHub Changelog: Copilot · Original published September 23, 2026
GitHub added local sandboxing to the Copilot app on September 23, 2026. It applies to sessions that work on a local repository or working tree, and it restricts the files, network resources, and credentials an agent's commands can reach. The point is to limit the damage from a command nobody intended to run.
What a project sandbox controls
Each project gets its own policy in three parts. For the filesystem, you set extra read-write folders, extra read-only folders, and denied folders. For the network, you allow or block outbound internet and local network access. For credentials, you decide whether Git credentials for HTTPS operations and GitHub CLI credentials are available inside the session.
If the operating system cannot enforce the requested policy, the sandboxed shell stops with an error instead of running without a sandbox. Enterprise-managed settings can make the effective policy stricter than the project's own.
Why the default matters
Sandboxing is off by default. You turn it on under the project's settings with Sandbox new sessions, and it applies only to sessions started after that. For a session already running, /sandbox on enables it without changing the project default.
The scope has limits worth knowing before you rely on it. Cloud sandbox sessions and sessions on a remote host are not covered. The Copilot app and Copilot CLI keep separate sandbox settings, so enabling one does not protect the other. GitHub labels the feature a public preview that may change.
Agent traces for administrators
Since September 22, 2026, the app can export agent activity through OpenTelemetry, an open-source observability framework. Administrators enable it with the telemetry property in the enterprise's managed-settings.json and point it at an endpoint. The traces follow a session step by step, including requests to models and the tools the agent used. Prompt and response content is excluded unless an administrator turns content capture on.
Code review settings on every plan
Personal settings for Copilot code review used to exist only on Copilot Pro, Pro+, and Max. As of September 23, 2026, a dedicated code review settings page is available on every plan, including Copilot Business and Copilot Enterprise, and the change is generally available.
- Automatic review can run when you open a pull request, coauthor one, or move one out of draft.
- Separate switches cover new pushes and draft pull requests.
- The default review effort is Lite or Balanced, and you can still pick another level when requesting a review by hand.
- Enterprise administrators can set one default effort for organization-owned repositories, which organizations and repositories can override.
„The sandboxed shell fails with an error rather than running without a sandbox.“
Sources
Related
OpenClaw 2026.10.2 stops long tool output from slipping credentials past redaction
A bug in OpenClaw's secret masking let a credential through whenever it straddled the boundary between two slices of a long text. The fix ships in 2026.10.2 along with repaired updates from 2026.10.1 and a new compaction default for OpenAI and Anthropic API keys.
OpenClawverified

OpenClaw 2026.10.1 lets agents search installed skills and changes how sessions message each other
OpenClaw released version 2026.10.1 on October 10, 2026, its first stable release since 2026.9.7. Agents can now search the skills installed on a machine and read only the one a task needs, instead of carrying the whole library in the prompt. Opt-in MCP Apps run beside chat, and full backups can go to an external drive or Cloudflare R2. Three changes can break existing setups: session-to-session messaging, the old required argument, and Workboard data stored before July 2026.
OpenClawverified
Claude Code 2.1.294 fixes prompt hooks that let through the actions they were written to block
Anthropic released Claude Code 2.1.294 on October 8, 2026. It fixes prompt and agent hooks phrased as instructions, which could allow exactly the actions they were meant to stop. Version 2.1.295 followed the same day with an onFailure setting that makes a crashed or timed-out command or HTTP hook block the action. Releases 2.1.292 and 2.1.293 from earlier in the week closed several permission and sandbox gaps.
Claude Code Releases (GitHub)verified
