GitHub Copilot can approve PRs

GitHub’s September 1 product changes push Copilot further into work that many teams still reserve for human reviewers. The company said Claude Fable 5.1 is now generally available in GitHub Copilot for “long-horizon, autonomous coding and knowledge-work tasks,” and separately said Copilot code review can now tell users when a pull request is ready to approve and, in public preview, submit an approval itself. The two changes land after several months of GitHub updates that expanded Copilot’s role in code review, including effort levels, agent skills, MCP support and more administrative controls over how reviews run. GitHub has also said Copilot review settings and billing rules are changing later this month, underscoring that the review product is still being actively tuned. (github.blog) ### How far does the new approval feature actually go? GitHub said Copilot does not approve pull requests by default. When an administrator enables the setting, Copilot can submit an approval that counts toward a repository’s required-approvals rule, according to the company’s changelog. GitHub also drew a line between recommendation and approval. (github.blog) The company said an “approval assessment alone” does not satisfy merge requirements; the counted approval only happens when the approval feature is turned on and Copilot submits it. If new commits are pushed after Copilot approves, that approval is dismissed the same way a human reviewer’s approval would be. (github.blog) ### What is Claude Fable 5.1 being added for? Anthropic’s Claude Fable 5.1 is being positioned by GitHub as a model for longer-running work inside Copilot. GitHub said the model is designed for “long-horizon, autonomous coding and knowledge-work tasks” and performed well in the company’s internal testing on deep codebase research, feature development and complex agentic workflows. (github.blog) GitHub also said Fable 5.1 differs from other Claude models in Copilot because it “requires data retention by default” so Anthropic’s safety classifiers can operate. That detail matters for enterprise administrators already managing model policies across Copilot Business and Copilot Enterprise, where GitHub has been rolling out global model-policy enforcement through September 1. (github.blog) ### What has GitHub been changing around code review before this? GitHub’s August 28 billing update said the default review effort value for repositories and organizations using Copilot code review will shift to “Balanced” starting September 28, 2026. Earlier August and June updates added effort levels, support for bot-authored pull requests under certain policies, and agent skills and MCP support to bring outside organizational context into reviews. (github.blog) GitHub has also reversed some earlier automation choices. On August 7, the company said GitHub Code Quality would no longer automatically add Copilot as a reviewer, saying teams had told GitHub that adding a reviewer should be their choice. ### Why does this raise a control question for teams? GitHub’s own product notes show that approvals, model access and review depth are now administrative decisions rather than fixed defaults. (github.blog) Repository and organization administrators can decide whether Copilot merely comments, signals readiness to approve, or submits an approval that counts toward merge rules. (github.blog) The security backdrop is broader than GitHub’s announcements. A separate security report cited in the source briefing said malicious `.git` configurations can induce command-line coding agents to run attacker-controlled code, including outside sandboxes in some cases. I could not independently retrieve that report from the web tool, so I cannot verify its specific findings here; the GitHub materials themselves do show the company adding more controls, runner configuration options and content exclusions around agentic review workflows. (github.blog) ### What happens next in GitHub’s rollout? September 28 is the next dated change GitHub has published for Copilot code review: the default effort level becomes Balanced for existing and new repositories and organizations using the product. The approval capability remains in public preview, while Fable 5.1 is listed as generally available in GitHub Copilot as of September 1. (github.blog 1) (github.blog 2)