What happened
GitHub’s release notes list Copilot CLI 1.0.109 as the latest release, dated November 10, 2026. Two additions are especially relevant to teams using the CLI as an agent runner. First, Claude Fable 5.2 is available in the model picker under “New models.” Second, sandbox host allow and deny rules in settings now support IPv4 and IPv6 CIDR ranges. That makes it possible to express network rules by address range rather than only as individual hosts, although the release notes do not describe the full rule syntax or precedence.
The release also includes two changes that affect automation boundaries. MCP tool results returned as a base64 blob resource with a text MIME type, including text/plain or application/json, now reach the model instead of arriving empty. In prompt mode, --output-format json emits a final result record with an exit status even when a run fails early. The latter gives downstream scripts a result marker and status to inspect in a failure path, rather than having to infer that no final record means an early exit.
Other listed fixes span terminal interaction, configuration and agent operation. Copilot now reports an unavailable selected or saved model instead of silently switching to another one; the release also improves rate-limit error wording, permissions for parallel file operations, and file-move prompts. It corrects settings descriptions and behavior for provider environment variables, MCP server removal, /init, /cd, and directory selection. The long list includes fixes to session history, subagent display, shell-mode navigation, and sandbox settings. Version 1.0.109 is available as downloadable builds for macOS, Linux and Windows, with multiple architectures represented in the release assets.
What the archive adds
New Runtime’s earlier coverage on Claude Fable 5.2 in GitHub Copilot noted its general availability for paid individual and business plans across Copilot clients on November 10, 2026. This CLI release is a new fact on that story: Fable 5.2 is now specifically listed in the Copilot CLI model picker. It is not evidence of a new model launch or a broader change to Copilot access.
Trend signal
This release weakly supports agent security moves to runtime boundaries: CIDR allow and deny rules put network controls in the sandbox configuration rather than relying on an agent prompt. It is a concrete configuration capability, but the release notes do not demonstrate how policies behave under adversarial use or how the controls are enforced in every execution environment.
Why now
The release follows GitHub’s November 10 announcement that Claude Fable 5.2 was generally available in Copilot for paid individual and business plans. Adding it to the CLI picker extends that model choice to another Copilot surface; the release does not frame the timing as a new general-availability event.
Why it matters
A model picker change matters when teams need to compare or select models within an existing agent workflow, but the more operationally significant update may be the sandbox’s address-range rules. Teams can define network access at a level suited to infrastructure ranges, which can make controlled execution easier to express than a growing list of individual hosts. The release does not establish whether this is sufficient for a particular organization’s security requirements, so it should be tested against the actual destinations a coding agent needs.
The MCP and JSON fixes matter at integration seams. A text result that reaches the model can restore an otherwise broken tool workflow, while a final JSON record with an exit status can make failure handling more explicit for scripts and job runners. Together, the changes are reminders to test both successful and failed runs when embedding a coding agent in CI or another managed process. They improve workflow reliability without claiming to change the CLI’s overall autonomy or guarantees.
What happens next
- Teams using Copilot CLI will add CIDR-based sandbox rules to network policy configurations. (3 months, 60%). Watch for: subsequent Copilot CLI settings or documentation examples using CIDR allow and deny entries.
Limits and uncertainty
GitHub’s release notes establish that CIDR support and the listed fixes ship in version 1.0.109, but they do not quantify adoption, security outcomes, or reliability improvements. They also do not provide a comparative evaluation of Claude Fable 5.2 in the CLI. The notes do not establish whether every listed sandbox behavior is identical across operating systems or execution modes, so teams should verify their own configuration. The release is a feature and maintenance update, not evidence that a particular network policy is sufficient for a given threat model.
Reader takeaway
Copilot CLI 1.0.109 adds Fable 5.2 as a model choice, but its more actionable operational change is CIDR-based sandbox host filtering. If you run CLI agents in managed environments, review the new network rules and test both MCP text results and failed prompt-mode JSON runs before relying on them in automation.