The report
GitHub made AI Scan for pull requests generally available on October 15, 2026, expanding security review to changes that ordinary CodeQL rules do not cover. The important permission boundary is easy to miss: an AI Scan finding is advisory. It does not fail the code-scanning check and cannot be used by this feature to block a merge through a ruleset requirement.
GitHub lists PHP, Shell/Bash, Terraform configuration, Dockerfiles, JSP for Java and Blazor for C# among the supported targets. The scan runs when code scanning and the effective repository AI Scan setting are enabled. It excludes pull requests from forks and those generated by Dependabot. Public repositories have general availability, while eligible private and internal organization repositories require GitHub Code Security or Advanced Security. The feature consumes Code Security AI Credits; a Copilot license is not required.
A review signal is not a release gate
A flagged change is a reason to investigate, not an independently reproduced exploit. A clean result does not show that every exploit path was tested or that an agent-generated fix preserves the original program's intended behavior. Existing P2Patch and PatchBench research demonstrates, under different evaluation methods, why blocking an initial proof of concept differs from fixing the underlying weakness and preserving functionality. Neither study measures the new GitHub service.
The GitHub release describes a workflow where findings appear alongside pull-request changes and may include suggested autofixes. That makes review information easier to surface but does not transfer approval authority to the scanner. An independent merge gate still requires a protected policy, reviewed code identity and evidence against tests the task agent could not simply rewrite. GitHub supplies no independently validated detection recall, false-positive rate or measured incident reduction in this announcement.