GitHub Copilot code review adds Adaptive effort and scans for 500+ secret types

GitHub has added three public preview features to GitHub Copilot code review and moved its Azure DevOps integration to general availability. In a changelog entry dated November 17, 2026, the company said Copilot can now pick its own review effort for each pull request with a new Adaptive setting, flag exposed credentials directly in review comments, and read related public repositories to find code that depends on a change. Adaptive effort, secret detection and cross-repository context are available on Copilot Pro, Pro+, Max, Business and Enterprise.
The update arrives three weeks after GitHub introduced the Max effort level, which runs several high-reasoning models on a single pull request. With Lite, Balanced and Max all in place, the open problem for teams was choosing among them, because each level costs a different amount of GitHub AI credits per review. Adaptive hands that choice to Copilot: it assesses the complexity and risk of each pull request, reviews it at Lite, Balanced or Max, and shows on the review which level it selected.
Secret detection combines two methods. For more than 500 known secret types, Copilot uses deterministic matching such as regular expressions, backed by the secret scanning partner program that GitHub runs with more than 150 technology partners. For credentials that follow no recognizable format, it uses large language models and the purpose-built secret detection model that GitHub introduced in October. The same changelog also adds review outcome reporting to Repository insights and makes Copilot code review for Azure DevOps generally available, with usage billed in GitHub AI credits to the linked Azure subscription.
··········
GITHUB COPILOT CODE REVIEW NOVEMBER 17 UPDATE AT A GLANCE.
The five changes have different release states and different scopes. The table below summarizes them as GitHub describes them in the changelog and in its code review documentation.
........
Feature | Status | Plans | Scope |
Adaptive review effort | Public preview | Pro, Pro+, Max, Business, Enterprise | Per PR or as a default at personal, repo, org or enterprise level |
Secret detection in reviews | Public preview | Pro, Pro+, Max, Business, Enterprise | Enabled in repo or org settings under Copilot, Code review |
Cross-repository context | Public preview | Pro, Pro+, Max, Business, Enterprise | Public repositories only |
Review outcome view in Repository insights | Available | Not stated | Any repository you can read |
Code review for Azure DevOps | Generally available | Not stated | Azure Repos, billed to the linked Azure subscription |
........
Balanced remains the default effort level for new and existing repositories, a setting GitHub has applied since September 28, 2026. Adaptive does not replace it automatically. Teams that want Copilot to choose the effort have to select Adaptive for a single pull request from the Reviewers bar or set it as the default in personal, repository, organization or enterprise settings.
··········
HOW ADAPTIVE EFFORT CHOOSES BETWEEN LITE, BALANCED AND MAX.
GitHub's documentation describes the three fixed levels in some detail. Lite is the standard review, aimed at fast feedback on common problems such as bugs, security vulnerabilities and style inconsistencies. Balanced routes the pull request to a higher-reasoning model for a longer analysis of complex logic, security-sensitive code and changes that cross service boundaries. Max uses several high-reasoning models on the same pull request and is intended for changes that need the most scrutiny before merge.
Adaptive sits above these three levels. According to the changelog, Copilot evaluates the complexity and risk of each pull request and then runs the review at the level it considers appropriate, labelling the result so that the author and reviewers can see which level was used. GitHub has not published the criteria Adaptive uses to score complexity or risk, so teams cannot yet predict in advance which pull requests will be escalated to Max.
The existing caps still apply. Organizations and enterprises can set a maximum review effort of No limit, Lite, Balanced or Max, and repositories can select that level or a lower one. GitHub states that Adaptive never exceeds the maximum set at the enterprise, organization or repository level. An enterprise that has capped reviews at Balanced will therefore see Adaptive choose only between Lite and Balanced, and no review will run at Max.
··········
WHAT ADAPTIVE EFFORT COULD MEAN FOR COPILOT CODE REVIEW COSTS.
Each Copilot code review is billed in two parts. Token consumption is charged in GitHub AI credits, where one credit equals $0.01, and the agentic infrastructure consumes GitHub Actions minutes, which are not included in the credit estimates. GitHub selects the model automatically and does not disclose it, so the cost of each review varies. The company publishes estimated ranges per effort level, and the table below reproduces them.
........
Effort level | Typical cost per review | Approximate average |
Lite | $0.05 to $1 | $0.25 |
Balanced | $0.25 to $5 | $1.50 |
Max | $1.50 to $25 | $10 |
........
GitHub labels these figures as estimates and says actual costs can fall outside the ranges, with consumption rising as pull requests grow and as repositories add custom instructions. Using the published averages, a team that opens 400 pull requests a month would spend about $100 in AI credits if every review ran at Lite, about $600 at Balanced and about $4,000 at Max.
Adaptive changes that arithmetic depending on how Copilot distributes the reviews, and GitHub has published no distribution. As an illustration only, if Adaptive sent 40% of those 400 pull requests to Lite, 50% to Balanced and 10% to Max, the monthly total at average prices would be about $740, or $140 more than reviewing everything at Balanced. If the same organization capped reviews at Balanced and Adaptive split the work evenly between Lite and Balanced, the total would fall to about $350, or $250 less than the current default.
The two scenarios show where the risk lies for budget owners. Adaptive can lower spending on routine pull requests, but a small share of reviews escalated to Max can outweigh the savings, because a single average Max review costs as much as 40 average Lite reviews. Teams that want to try Adaptive without exposure to Max pricing can set the maximum review effort to Balanced first, and GitHub's budgets can block reviews once a limit is reached.
Billing attribution follows the existing rules. Automatic reviews are usually charged to the pull request author and manual reviews to the person who requested them, depending on the organization's billing policy. Reviews of pull requests opened by Copilot cloud agent go first to the human co-author, and reviews of pull requests opened or requested by other bots go to the organization. Usage can be traced in billing reports by filtering on the workflow path dynamic/agents/copilot-pull-request-reviewer.
··········
[[ADPLUS_300x250_1]]
SECRET DETECTION IN COPILOT CODE REVIEW AND HOW TO TURN IT ON.
Secret detection runs inside the review itself. When Copilot finds an exposed credential in a pull request, it raises it in a review comment so that the author can remove it before the change is merged. GitHub says the deterministic layer covers more than 500 secret types, while the model-based layer looks for passwords and other credentials that have no recognizable token format.
The model-based layer draws on the fine-tuned secret detection model GitHub announced on October 7, 2026. That model reads the code around a string to judge whether it is likely to be a credential, and GitHub said at the time that it does not generate code or prose. The company has not published precision, recall or false-positive figures for the model, either in October or in the November changelog.
To enable the feature, a repository or organization administrator opens the settings, selects Copilot and then Code review. The changelog does not say whether secret findings in reviews consume additional AI credits beyond the review itself. The feature also operates separately from GitHub Secret Protection: push protection and secret scanning alerts remain part of that product, and the October changelog stated that AI push protection on GitHub.com requires paid Secret Protection or Advanced Security coverage.
··········
COPILOT CROSS-REPOSITORY CONTEXT AND THE REPOSITORY INSIGHTS REVIEW VIEW.
Cross-repository context lets Copilot read related repositories while reviewing a pull request. GitHub says the reviewer can find code in other repositories that depends on the change under review, and can also use documentation, custom instructions and skills stored in those repositories. In the current preview this works with public repositories only, which limits its value for companies whose services and shared libraries sit in private repositories.
The review outcome view in Repository insights adds reporting. Under the Insights tab, in Repository insights, anyone who can read a repository can now see how Copilot code review feedback is addressed and used in that repository. The view complements the addressed rate that GitHub added to its usage metrics API earlier in November, which we covered in our report on the Copilot code review admin APIs and addressed rate.
··········
COPILOT CODE REVIEW FOR AZURE DEVOPS GENERAL AVAILABILITY AND BILLING.
Copilot code review for Azure Repos had been in public preview. With general availability, administrators can enable it at the organization, project or repository level in Azure DevOps. A branch policy can request a Copilot review automatically on every pull request, or users can add Copilot from the Reviewers list on a single pull request.
The Azure DevOps version supports a default effort of Lite or Balanced for each project or repository, with overrides on individual reviews. The changelog does not mention Max or Adaptive for Azure DevOps. Teams can add custom instructions and agent skills, apply and commit suggestions while resolving the related comments in the same step, and audit enabling, disabling and settings changes through audit log events.
Billing works differently from GitHub.com. Usage is charged in GitHub AI credits to the Azure subscription linked to the Azure DevOps organization, and costs can be tracked with budget alerts in Azure Cost Management. For companies that already buy Azure services, this puts code review spending on the same invoice as the rest of their cloud usage instead of a separate GitHub bill. GitHub has not stated which Copilot plans are required for the Azure DevOps integration.
··········
WHAT CHANGES FOR GITHUB COPILOT USERS AND ADMINISTRATORS.
Individual developers on Copilot Pro, Pro+ or Max can try Adaptive on their own pull requests from the Reviewers bar or set it as their personal default, which applies to automatic reviews and to reviews they request manually. Because individual plans include a limited allowance of AI credits that varies by plan, developers who review frequently will want to watch whether Adaptive draws down that allowance faster than Balanced. The differences between the individual allowances are set out in our comparison of GitHub Copilot Pro and Pro+.
Administrators on Copilot Business and Enterprise have more decisions to make. The first is whether to set Adaptive as the organization default or leave Balanced in place while a few repositories test it. The second is the maximum review effort, which controls whether Adaptive can reach Max at all. The third is whether to enable secret detection in reviews for repositories that do not already have Secret Protection. Business and Enterprise plans pool per-user AI credit allowances at the billing entity level, so a change in review effort across many repositories affects a shared budget. Plan-level differences are covered in our comparison of GitHub Copilot Business and Enterprise, and the Max level itself in our report on Max multi-agent reviews.
All three new GitHub.com features are previews, and GitHub states that preview features are subject to change. Organizations that depend on predictable review costs can keep their current configuration unchanged, since nothing in the update alters existing defaults.
··········
GITHUB COPILOT CODE REVIEW ADAPTIVE EFFORT: CONFIRMED FACTS AND OPEN QUESTIONS.
GitHub has confirmed that Adaptive chooses among Lite, Balanced and Max for each pull request, labels the level it used, respects every effort cap and is available as a public preview on all paid Copilot plans. It has confirmed that secret detection covers more than 500 known secret types and uses a dedicated model for unformatted credentials, that cross-repository context is limited to public repositories, and that Azure DevOps reviews are generally available and billed through Azure.
Several points remain undisclosed. GitHub has not explained how Adaptive scores complexity and risk, how often it selects each level, or whether it raises or lowers average spending compared with Balanced. It has not published accuracy figures for the secret detection model, has not said whether secret findings carry their own credit cost, and has not listed the plans required for Azure DevOps. Until GitHub publishes data on how Adaptive distributes reviews, the cautious configuration for cost-sensitive teams is Adaptive with a Balanced cap, which keeps the per-review ceiling at the level they already pay for today.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org




