Skip to content
NewsProduct launch

GitHub's AI Scan now explores whole repositories, not just pull request diffs

· by Pondero Newsdesk

The short version

GitHub's agentic AI Scan can now crawl an entire repository and flag vulnerabilities in languages CodeQL never covered, a scope jump from the pull-request-only version it shipped in September.

GitHub's AI Scan now explores whole repositories, not just pull request diffs

GitHub's agentic vulnerability scanner no longer stops at the lines a pull request changes. As of October 28, per GitHub's changelog, AI Scan for repositories can crawl an entire codebase and surface issues in languages CodeQL has never supported, a scope jump from the diff-only version GitHub shipped in September.

What shipped

AI Scan for repositories entered public preview on github.com and GitHub Enterprise Cloud with data residency, per the changelog. Once an enterprise enables the AI Scan policy, organization owners and security managers turn it on for their org; personal-account owners flip it on in repository settings. A scan analyzes the latest commit on the default branch, and GitHub describes the underlying tool as an agentic code scanner that explores a repository and focuses on the code most likely to hold vulnerabilities, running within a fixed time budget to keep cost bounded. Findings land in the normal code scanning alert list next to CodeQL alerts, support assignees and linked pull requests, and can be fixed with agentic autofix. Where AI Scan and CodeQL cover the same repository, AI Scan skips what CodeQL already reports. The feature does not require a GitHub Copilot license, and GitHub separately confirmed the same day that its pull-request-scoped sibling, "Code scanning for AI applications," which flags indirect prompt injection and excessive agency, stayed in public preview as a distinct, narrower product.

A scan budget replaces the Copilot-license gate, but usage still meters

The access model here breaks from how GitHub has gated most AI security tooling this year. Public repositories, whether org-owned or on a personal account, get AI Scan for repositories without a Copilot seat attached. Private and internal repositories still need GitHub Code Security or GitHub Advanced Security, and personal accounts cannot run it on private repos at all. That makes the real constraint billing, not licensing: usage draws down Code Security AI credits, shows up under the Advanced Security product in enterprise billing, and GitHub will warn before a scan starts if a team has burned through its SKU-level budget. Rescans also do not auto-close fixed alerts, so a security team has to kick off a fresh scan after merging a fix to clear its own backlog, a manual step easy to skip under deadline pressure.

Context

GitHub built this on the same AI Scan infrastructure it opened up programmatically with the AI Scan for pull request APIs preview in September. That release let organizations enable AI-powered detections across repos via API; this one extends the detection surface from a pull request's changed lines to the entire default branch. Pondero covered the companion "Code scanning for AI applications" launch, which targets prompt-injection-shaped risk in agent code on pull requests, in a separate piece on October 29. The two previews shipped the same day at GitHub Universe but solve different problems: one watches what changes, the other audits what already exists.

What to watch next

GitHub has not published false-positive rates or example findings from AI Scan for repositories, and the credit-metering model means cost at scale is still unclear for teams running it broadly rather than on a handful of flagship repos. Watch for whether GitHub extends automatic alert-closing to rescans, and whether the policy-gated enterprise rollout widens to GitHub Enterprise Server the way the pull-request version eventually should.

Sources