GitHub AI Scan is now generally available — but its findings cannot block merges

GitHub made [AI Scan for pull requests generally available](https://github.blog/changelog/2026‑10‑15‑ai‑scan‑for‑pull‑requests‑generally‑available/) on October 15. It examines changed code for vulnerability patterns in languages and frameworks that CodeQL does not currently cover, including PHP, Shell/Bash, Terraform HCL, Dockerfiles, JavaServer Pages and Blazor. That makes the feature a coverage layer rather than a replacement for CodeQL. GitHub says AI Scan runs when code scanning is enabled and the repository’s effective AI Scan setting is on. Its findings appear beside CodeQL results in the pull request with an AI label, an explanation of the risk and, when available, a suggested fix. Teams should keep the two result types distinct in triage. CodeQL is a query‑based analyzer with established rules and configuration; AI Scan adds model‑generated detections for changed code. A useful rollout records which engine produced each alert, how reviewers resolved it and whether the same issue appears in later scans.

The central operational limit is explicit: AI Scan findings do not fail the code‑scanning check and cannot block a pull request through ruleset requirements. A repository can therefore show an AI finding while every required check still passes. That matters for teams that treat a green branch‑protection result as a complete security decision. If an AI Scan finding needs mandatory review, the process has to supply that control through ownership, triage policy or a separate approved workflow. GitHub’s [AI‑powered detections documentation](https://docs.github.com/enterprise‑cloud@latest/code‑security/concepts/code‑scanning/ai‑powered‑security‑detections) should be the configuration reference; the release notice is not evidence that a repository’s policy actually enforces review. Start with a visible queue and a response‑time target. Measure confirmed findings, dismissed findings and unresolved age separately. Do not convert a model score into severity without examining reachability, data flow and the affected deployment context.

GitHub says AI Scan does not run on pull requests from forks or on pull requests opened by Dependabot. That creates a predictable coverage boundary for open‑source contribution flows and dependency updates. Repository owners should map those exclusions before claiming complete pull‑request coverage. Keep existing CodeQL, dependency review, secret scanning and human review controls on the excluded paths. For external contributions, test the code after it lands on a trusted branch or through the project’s established safe validation process rather than granting untrusted workflows broader credentials merely to obtain an AI scan. The exclusion also affects metrics: an AI Scan completion rate calculated only over eligible pull requests can look healthy while leaving a substantial share of changes outside the feature. Report the eligible denominator and the excluded pull‑request classes.

AI Scan is available for public repositories owned by organizations or personal accounts. Private and internal repositories owned by organizations require GitHub Code Security or GitHub Advanced Security. GitHub says a Copilot license is not required. The entitlement and the consumption meter are separate. GitHub’s release says AI Scan uses Code Security AI Credits, appears under the Advanced Security product in enterprise AI usage and supports SKU‑level budgets. Its [Advanced Security billing documentation](https://docs.github.com/billing/concepts/product‑billing/github‑advanced‑security#how‑ai‑feature‑usage‑is‑billed) is the current source for how those credits are charged. Before enabling the feature widely, set a budget owner and alert threshold, then sample cost per eligible pull request. A successful pilot should show both security value and a cost profile; alert volume by itself proves neither.

Use a small representative repository set first: one language CodeQL already covers, one newly covered language or configuration format, and one workflow with fork or Dependabot activity. Confirm the effective setting, trigger an ordinary test pull request, identify the AI‑labelled result and verify that branch protection behaves exactly as documented. Then review findings with application context. Track true positives, useful‑but‑nonsecurity observations, dismissals and time to resolution. Preserve a rollback path for budget or noise problems, and make the advisory status visible to reviewers so nobody mistakes a green check for an empty queue. The release is useful because it adds a new source of evidence on pull‑request changes. Its limits are equally important: it is not comprehensive, does not enforce a merge gate and does not remove the need for existing scanners or review. Follow further platform‑security changes on the [GitHub organization page](https://opentools.ai/organizations/github). *Image: GitHub’s official screenshot for the October 15 AI Scan release, reproduced from the source page without generative alteration or artificial upscaling.*