GitHub’s September 25, 2026 changelog covering the week of September 21 brings several Copilot changes across models, local development, observability, collaboration tools, JetBrains IDEs, and VS Code. The most consequential updates are broader model availability, public-preview local sandboxing in the Copilot app, and OpenTelemetry support for agent monitoring.
The release also changes how developers can use Copilot cloud agent from Slack and Microsoft Teams, adds new JetBrains workflows, and extends VS Code agent sessions into remote Dev Containers. Availability varies by Copilot plan, client, platform, and rollout status, so the announcements should not be read as universal feature availability.
What changed in GitHub Copilot
Four models are listed as available
The changelog lists four newly available models with different plan eligibility:
- Claude Opus 5.5: Copilot Pro+, Max, Business, and Enterprise.
- GPT-6 Sol: Copilot Pro+, Max, Business, and Enterprise.
- GPT-6 Luna: Copilot Pro, Pro+, Max, Business, and Enterprise.
- Grok 4.7: Copilot Pro, Pro+, Max, Business, and Enterprise.
GitHub’s supported-models reference also lists these models and emphasizes that availability depends on both the Copilot plan and the surface where Copilot is used. Model availability can change over time. The supplied documentation does not establish regional, account-level, seat-level, or universal client availability for each model.
For teams evaluating a model change, the practical takeaway is to verify availability in the specific Copilot client and plan used by the affected developers rather than assuming that plan eligibility guarantees access everywhere.
Local sandboxing arrives in public preview
The Copilot app now offers local sandboxing in public preview. The feature is intended to limit an agent’s access to local files, networks, and credentials while it works in a local repository or working tree.
Sandbox settings are configured separately for each project. Administrators or developers can control additional or denied filesystem paths, outbound internet access, local-network access, and whether Git or GitHub CLI credentials are available inside the sandbox. The settings apply to local repository and working-tree sessions, not cloud-sandbox or remote-host sessions.
Local sandboxing is off by default for Copilot app projects unless enterprise-managed settings require it. Policy changes apply to new or restarted sessions rather than automatically changing an already-running session. That distinction matters when testing a policy: restarting the session may be necessary before its restrictions take effect.
Platform requirements also apply. The documented Linux prerequisites include Bubblewrap 0.5.0 or later and slirp4netns. Windows requires the supported Windows 11 versions and updates described in the local sandbox configuration documentation. If a host cannot enforce a requested policy, commands may fail rather than run without the intended restriction.
OpenTelemetry adds agent monitoring
The Copilot app can send agent-monitoring data through OpenTelemetry to an OTLP-compatible backend or collector. The monitoring data can include traces, metrics, and events about agent sessions, model calls, tool use, token usage, and actions such as edit feedback.
Setting up the integration requires an observability backend or collector, along with client configuration for the endpoint, headers, and authentication token. Enterprise-managed settings can enforce the telemetry configuration across supported clients. The OpenTelemetry documentation describes the available data and configuration requirements.
There is an important data-handling boundary. Prompts, responses, and tool arguments are excluded by default. Administrators can choose to capture that content, but doing so may send code, file contents, and user prompts to the configured observability backend. Organizations should therefore review retention, access, and data-handling practices before enabling optional content capture for sensitive repositories.
Slack and Teams become more useful—and more consequential
The weekly changelog reports updates for Copilot cloud agent in Slack and Microsoft Teams, including model selection and switching, Slack channel model defaults, issue-duplication checks, clearer implementation-plan status, and improved recovery or reconnection behavior.
Both integrations are documented as public-preview features for paid Copilot plans. They require the relevant Slack or Teams account, the GitHub integration or app installation, and cloud sandboxes enabled for the applicable Copilot plan. Organization or enterprise members may also need an owner to enable cloud agent or cloud sandboxes.
The Teams integration documentation provides additional operational details. From Teams, Copilot cloud agent can investigate, plan, write code, and create issues or pull requests using conversation context. Tasks run asynchronously in a secure cloud sandbox and post results when ready.
Teams captures the entire thread as context and stores that context in generated artifacts. A direct message can be used to limit the context. In direct messages, actions use the linked GitHub personal account’s permissions. In shared contexts, artifacts such as pull requests use the Copilot app identity. Only users with repository write access can trigger Copilot to make changes through Teams; other participants can provide input, while guest workspace members and outside repository collaborators cannot start or steer a session.
These context and attribution rules should be part of a team’s review process. Avoid placing unnecessary sensitive information in shared threads, and review repository rulesets when pull requests are created under the Copilot app identity. A shared-context pull request requires one additional approval before merging when repository rulesets already require at least one approval.
JetBrains and VS Code updates
JetBrains users get assisted approvals in public preview, earlier-message editing with conversation and file-change rewind, and support for shared organization or enterprise skills and managed custom instructions in local and Copilot agent sessions. The changelog announces these capabilities, but the supplied documentation does not provide detailed assisted-approval risk classifications or a complete workflow description. Teams should treat the exact behavior as preview functionality subject to change.
VS Code 1.139, released September 23, 2026, adds a more specific remote-development capability: agents can run inside Dev Containers for projects hosted on SSH, Tunnel, and WSL environments. This lets the agent use the remote project’s tools and dependencies.
Remote Dev Container agent sessions require a supported Dev Container configuration and Docker on the remote host. The feature is rolling out gradually and can be enabled with chat.agentHost.devContainer.enabled when available, or through the Agents Window’s Use Dev Container action. VS Code 1.139 also adds Compact View, filtering for empty session groups, inline renaming for sessions and nested chats, and a preview choice between multiple chat tabs and a single active-chat view while preserving conversation state.
What developers and administrators should do
- Check model access in context. Confirm the plan and Copilot surface used by each team before standardizing on one of the newly listed models. The published plan lists do not guarantee regional or universal client availability.
- Test sandbox policies on representative projects. Configure filesystem, network, and credential restrictions per project, verify the host prerequisites, and start or restart sessions after policy changes.
- Design telemetry controls deliberately. Prepare an OTLP-compatible backend or collector and review endpoint, header, and token configuration. Keep optional prompt, response, and tool-argument capture disabled unless its data-handling implications are acceptable.
- Review collaboration permissions. Before enabling Slack or Teams code-changing workflows, confirm paid-plan access, app installation, cloud-sandbox enablement, repository write access, and repository approval rules.
- Control conversation context. Use direct messages when a full shared thread is unnecessary, since Teams can include the entire thread in agent context and generated artifacts.
- Validate remote VS Code prerequisites. Confirm the remote host has Docker and a supported Dev Container configuration before adopting the gradually rolling-out agent support.
Availability caveats
Several changes are previews or gradual rollouts: local sandboxing in the Copilot app, JetBrains assisted approvals, and Slack and Teams cloud-agent integrations are public preview, while VS Code remote Dev Container agent support is rolling out gradually. The release notes also do not document every implementation detail for Slack, Teams, or JetBrains, and they do not establish performance, latency, billing, rate-limit, or model-specific limits.
For teams adopting these changes, the safest approach is to validate the exact plan, client, platform, permissions, and managed policies in the target environment before making the features part of a standard development workflow.
