News

GitHub Lets Security Teams Start AI Scans Across 20 Repositories, but Findings Still Need Review

GitHub now lets teams start AI security scans on up to 20 repositories at once. The preview still requires manual review and consumes AI credits.

GitHub is letting security teams send its AI vulnerability hunter into up to 20 repositories at once. The November 16 update cuts the work of launching scans across a codebase spread over multiple projects, but leaves people responsible for judging what the machine finds.

The new bulk-start option is in public preview. Teams can launch scans while enabling the organization setting or from the organization's code scanning alerts page. An individual repository can still be scanned separately.

AI Scan checks a repository's default branch and puts its findings alongside CodeQL alerts, the existing code analysis results. GitHub says it can extend coverage to languages CodeQL does not handle and identify additional issues in supported languages. A Copilot license is not required.

A bigger launch button, with limits

GitHub's documentation describes an agent that chooses files and functions to inspect within a time budget. It does not read every file exhaustively. Findings may be wrong and need assessment before anyone acts on them.

The scans run on demand. Scheduled runs and automatic scans on push are unsupported, and merging a proposed fix does not automatically start another scan during the preview. Someone must launch a fresh check.

Each scan consumes Code Security AI credits. Private or internal repositories owned by organizations require GitHub Code Security or GitHub Advanced Security; private repositories owned by personal accounts are unavailable. Bulk access therefore does not mean unrestricted scanning for every project.

Enabling the feature is a separate step

The release follows GitHub's November 13 REST API update, which lets organizations manage the feature's enabled state through scripts. That setting alone does not start a scan, and enterprise policy can block enablement.

For a security team with a long backlog, today's change removes repeated launch work. The resulting alerts still land in a queue that people have to investigate.

Some offers on this page may be paid placements or contain affiliate links.

Cyberpresso: daily cyber & AI brief

Free daily newsletter, read in 5 minutes.

Subscribe free