GitHub Copilot introduced a set of new AI models, a public‑preview sandbox for the desktop app, and a batch of integration refinements across Slack, Microsoft Teams, JetBrains IDEs, and VS Code. These changes affect how AI‑assisted coding, issue creation, and tool‑calling are provisioned, monitored, and secured, which is directly relevant to AI engineers, platform teams, SREs, and security practitioners.
Copilot model updates and plan eligibility
The weekly release adds four models to the Copilot catalog:
- Claude Opus 5.5 – available to Pro+, Max, Business, and Enterprise subscriptions.
- GPT‑6 Sol – same plan coverage as Claude Opus 5.5.
- GPT‑6 Luna – available to Pro, Pro+, Max, Business, and Enterprise plans.
- Grok 4.7 – also available to Pro, Pro+, Max, Business, and Enterprise plans.
Teams can now select a model per task and switch mid‑conversation, with the choice persisting for the remainder of the thread. This granularity lets engineers match model capabilities to workload cost, latency, or compliance constraints.
Local sandboxing and OpenTelemetry in the Copilot app
The Copilot desktop client now offers a public‑preview sandbox that restricts an agent’s file system, network, and credential access. For security teams, this provides an explicit boundary that limits accidental credential leakage or unintended network calls. The sandbox can be observed via OpenTelemetry, configured through enterprise‑managed settings, allowing existing monitoring stacks to ingest agent activity without additional instrumentation.
Integration refinements for collaboration and IDE workflows
In Slack and Microsoft Teams, Copilot now checks for existing similar issues before opening a new one, reducing duplicate tickets. Users can share files, attachments, and message links to give the agent richer context. JetBrains IDEs gain a public‑preview “assisted approvals” flow: low‑risk tool calls are auto‑approved, while higher‑risk actions trigger a prompt. Editing a prior message rewinds the session state, effectively resetting both the conversation and any pending file changes. VS Code 1.139 adds the ability to run Copilot agents inside Dev Containers on SSH, Tunnel, and WSL hosts, leveraging the remote project’s toolchain. UI tweaks include a compact session view, filters for empty groups, and a preview mode that lets users choose between separate chat tabs or a single active‑chat view.
Related CloudNinjas coverage: AI engineering.
What This Means For Practitioners
Teams should audit which model tier aligns with their cost and compliance posture, and consider pinning a model for critical pipelines. The sandbox introduces a new security control surface; verify that sandbox policies match your organization’s credential‑handling standards and that OpenTelemetry endpoints are correctly scoped. For CI/CD and incident‑response workflows, test the Dev Container agent path and the assisted‑approval flow to ensure they integrate with existing approval gates. Finally, monitor the rollout of these features for stability and watch for any future extensions to the sandbox or telemetry schema that could affect observability pipelines.