Copilot code review now picks its own effort level and flags leaked secrets
GitHub's Copilot code review can now choose Lite, Balanced or Max for each pull request, and flag exposed secrets, in public preview on paid plans.

GitHub, part of Microsoft, updated Copilot code review on November 17. A new Adaptive mode picks Lite, Balanced or Max for each pull request, and reviews can now flag exposed secrets. Both are in public preview.
Adaptive decides how hard to look
Until now, a person had to choose the review effort level in settings or when requesting a review. Adaptive moves that choice to Copilot, which looks at each pull request’s complexity and risk. GitHub’s own example is that a documentation edit may need only Lite, while a change to authentication logic benefits from Max. Every review shows which level Adaptive picked.

You can select Adaptive for one pull request from the Reviewers bar, or make it the default in personal, repository, organization or enterprise settings. Balanced stays the default level. Adaptive is capped: it never goes above the maximum effort level set by the enterprise, organization or repository, so admins keep a ceiling on how deep, and presumably how costly, a review can get.
All three previews, Adaptive, cross-repository context and secret detection, are open to subscribers of the paid Copilot tiers, from Pro up to Enterprise. The announcement does not list the Free plan.

Secrets and other repositories enter the review
Secret detection uses two approaches. The first covers known formats: GitHub says it works with more than 150 technical partners through its secret scanning partnership program, and that this yields over 500 secret types found with deterministic methods such as regex. The second covers secrets with no recognizable format, which Copilot code review looks for with large language models and a GitHub model built for that purpose. Findings appear as review comments so the credential can be removed before merge. To turn it on, open repository or organization settings, choose Copilot, then Code review.
Cross-repository context lets Copilot read related public repositories during a review. That can surface code elsewhere that depends on the change, along with documentation, custom instructions and skills stored in other repositories. For now it applies to public repositories only, so private multi-repo codebases, where most enterprise dependencies live, are not covered by this preview.

GitHub also announced that Copilot code review for Azure Repos is now generally available. Admins can enable it at organization, project and repository level, add a branch policy to request reviews automatically, and set a default of Lite or Balanced. Usage is billed in GitHub AI credits to the Azure subscription linked to the Azure DevOps organization. Note that Adaptive is not listed among the Azure DevOps options in the announcement.
What to watch before trusting the automation
The announcement gives no accuracy figures, false positive rates or cost numbers for Adaptive or secret detection, so the claims are GitHub’s own and unverified. The useful test is whether Adaptive spends Max effort on the pull requests that deserve it, and whether it under-reviews changes that look routine but are not.
A companion post gives teams a way to check. A new Repository insights page, the first item in a repository’s Insights tab, tracks comments addressed, high-severity comments addressed, and pull requests with no later human change request. It also counts merged pull requests that Copilot never reviewed, and shows an addressed rate with a trend over time. Anyone with read access can see the page, according to GitHub, though the Copilot usage metrics policy must be enabled. That gives reviewers a measurable signal for whether Adaptive’s choices pay off.
The related MAI Code 2 post shows GitHub is also widening the model choices in Copilot, with MAI Code 2 for Pro and above and the cheaper MAI Code 2 Flash reaching Free and Student plans, but it says nothing about code review. It is not stated which models power Adaptive reviews.
The release fits a week of Copilot tooling news. Our coverage of Copilot Evals described a way to rerun skills and MCP tools, and the new review features rely on the same custom instructions and skills that teams are being encouraged to test. Because features in public preview are subject to change, teams with strict compliance rules should treat secret detection as an extra layer, not a replacement for existing secret scanning.
Sign Up for the Daily AI Pulse
One email a day. All the stories that matter.
By signing up, you agree to our Terms of Use and acknowledge the data practices in our Privacy Policy. You may unsubscribe at any time.
Leo MartinsAI Tools, Prompts & Practical AI ExpertLeo Martins is the AI Tools & Practical AI Expert at Aiholics, focused on helping readers use artificial intelligence to improve productivity, creativity and everyday work.
You may also like

Copilot can now test itself: Evals reruns your skills and MCP tools

Copilot CLI can now sign in with Entra, and bill Microsoft 365 instead of GitHub

Copilot app can now open a pull request in Azure DevOps and drive your browser
