GitHub AI Scan for Repositories: Preview, Billing and Limits

GitHub released AI Scan for repositories in public preview on October 28, 2026. Unlike the existing pull-request scanner, the new mode examines the latest commit on a repository’s default branch and can cover languages or frameworks outside CodeQL support. It is a separate setting, a manually started scan and a billed AI feature.

An agentic scanner explores within a fixed time budget

After an enterprise or repository policy enables the feature, a user with code-scanning write access can start a scan. GitHub says the scanner focuses on code most likely to contain vulnerabilities and stays within a fixed time budget. When CodeQL covers the same repository, AI Scan skips issues already reported by CodeQL.

A bounded scan can miss code it does not prioritize or cannot analyze in time. Findings are advisory evidence rather than a clean bill of health. Security teams should retain conventional static analysis, dependency scanning, testing and threat modelling.

Fixes require a new scan before alerts close

AI Scan findings enter the normal code-scanning backlog and support assignees, linked pull requests, issues, agentic autofix and security campaigns. GitHub says an alert does not automatically close when a fix is merged. A new scan must confirm the issue is gone.

That rule prevents a merge event from being mistaken for verification. Teams should review the proposed fix, run tests and rescan the exact default-branch commit. If the scanner does not rediscover a finding, investigate whether code changed, coverage shifted or the fix genuinely removed the path.

Availability differs by ownership and repository visibility

The preview is available on github.com and Enterprise Cloud with data residency. Organization-owned public repositories are eligible; private and internal repositories require GitHub Code Security or Advanced Security. Personal-account public repositories are eligible, while personal private repositories are not. A Copilot licence is not required.

These distinctions matter for Canadian teams choosing where code and findings are processed. Data-residency product availability does not by itself answer every contractual requirement; confirm the selected enterprise region and security terms.

AI credits and budgets make scan scope an operational choice

Repository scans consume Code Security AI credits and appear under Advanced Security AI usage. GitHub says enterprises can set SKU-level budgets and warns when the budget is exhausted before a scan starts. Public personal repositories may be billed to the owner or to an organization providing the owner’s Copilot licence.

Budget controls cap spend, not security risk. Define which repositories and events merit a scan, record cost per reviewed finding and avoid using scan frequency as a substitute for remediation. Public-preview behaviour and pricing may change, so keep the policy and cost basis dated.

Sources and editorial method

We use the linked original statements and reports to distinguish documented facts from our analysis. Company announcements describe their authors’ plans and claims; they are not independent verification of future outcomes. This article is dated and will be revised if material evidence changes. Read our editorial standards and corrections process.

GitHub: AI Scan for repositories in public preview ↗ — Official October 28 announcement; operation, eligibility, billing and rescan rules.