GitHub Brings AI Security Scanning to Pull Requests CodeQL Can’t Cover

GitHub has made AI Scan for pull requests generally available. The feature complements CodeQL by finding vulnerabilities in pull request changes for languages and frameworks that CodeQL does not currently cover, including PHP, Shell/Bash, Terraform configuration (HCL), Dockerfiles, JSP for Java, Blazor for C#, and more.

For DevOps teams, several of those files define how software gets built and deployed. A single pull request can touch a PHP controller, a deploy script, and a Terraform module, and now all three get a security review. Until now, CodeQL could only weigh in on part of that.

How It Works

AI Scan runs automatically on pull requests when code scanning and the repository’s AI Scan setting are enabled. Findings appear alongside CodeQL results on the Conversation and Files changed tabs, with an “AI” indicator to set them apart from CodeQL alerts. A finding can include an explanation of the risk and a suggested fix.

There’s an important limit. AI Scan findings are advisory. They don’t fail the code scanning check, and ruleset requirements can’t block merges. AI Scan also doesn’t run on pull requests from forks or on pull requests created by Dependabot. The fork exclusion matters for open source maintainers. Outside contributors to public projects usually open pull requests from a fork, so those changes get no AI Scan at all.

What Changed Since Preview

GitHub first put AI security detections on pull requests into public preview in July. At that stage, the feature required a GitHub Copilot license and drew down an organization’s AI credits. The GA release drops that requirement. AI Scan is available for public repositories owned by organizations or personal accounts. Private and internal repositories need GitHub Code Security or GitHub Advanced Security. A Copilot license isn’t required.

Billing has shifted as well. AI Scan consumes Code Security AI Credits, which show up in enterprise billing under the Advanced Security product. Teams can set SKU-level budgets for those credits. Neither the changelog nor the docs say how many credits a single pull request scan uses, so cost planning will take some observation.

GitHub has added management tools too. In September, it released REST API endpoints in public preview that let teams turn AI Scan on or off at the organization and repository levels. Repository settings can’t override an organization-level disabled state. On Oct. 13, GitHub also added AI Scan metrics to its security overview.

Paired With Agentic Autofix

GitHub shipped a related feature the same day. Agentic autofix for code scanning alerts is now generally available. It explores relevant files, proposes a fix, and reruns CodeQL to confirm the fix closes the alert before opening a pull request for review. It’s available to Code Security and Advanced Security customers on GitHub Enterprise Cloud, and it consumes AI Credits from the Code Security budget plus GitHub Actions minutes. It only supports alerts on the default branch.

Taken together, the two releases show where GitHub is heading. One tool flags problems in new code before it merges. The other fixes problems already in the codebase. Both still put a human at the review step.

“Moving detection into the pull request puts security closer to where code is written, which matters as agents write more of it,” said Mitch Ashley, vice president and practice lead for CIO & Technology Buyers and Software Lifecycle Engineering at The Futurum Group. “Advisory findings with no merge gate only reduce risk if a named reviewer owns them.”

Why It Matters Now

The timing makes sense. More code is coming from AI assistants and agents, and that code isn’t secure by default. Veracode’s 2025 GenAI Code Security Report tested more than 100 LLMs and found that 45% of AI-generated code samples failed security tests against the OWASP Top 10. Georgia Tech researchers tracking CVEs tied to AI-generated code counted at least 35 in March 2026, up from six in January.

Volume is the other problem. When pull requests pile up faster, reviewers can’t read every line with the same care. Automated checks on every change help close that gap. But they only help if they cover the risk-carrying files.

“Terraform modules, deploy scripts, and Dockerfiles define production and have long been scanner blind spots,” Ashley said.

A second detection engine has value on its own. One 2026 study of 522 AI-generated code samples found that only one SAST tool caught about 60% of confirmed vulnerabilities. Different scanners catch different things.

What DevOps Teams Should Consider

• Assign an owner. Since AI Scan can’t block merges, decide who reviews AI-labeled findings and how fast.
• Watch the credit burn. Set SKU-level budgets before turning the feature on across hundreds of repositories.
• Know the gaps. Fork-based contributions and Dependabot pull requests aren’t scanned. Open source projects need other controls for outside code.
• Roll out in stages. The API endpoints make it easier to start with repositories that carry the most IaC and scripting code.

The Bottom Line

AI Scan doesn’t replace CodeQL. It fills in coverage where CodeQL doesn’t reach, and those are files DevOps teams touch every day. Advisory-only is a reasonable starting point for an AI-based detector. But it puts the burden on teams to act on what it finds.

And accuracy will decide whether that works. “Watch signal quality,” Ashley said. “If AI findings arrive faster than reviewers can clear them, teams trade one backlog for another.”