GitHub AI security detections entered public preview on July 14, 2026. The new scanner posts AI-labeled security findings directly on pull requests and is designed to cover languages and frameworks that CodeQL does not currently analyze.

What changed

When an enabled repository opens or updates a pull request, GitHub’s AI detection engine analyzes the changed code and can use repository search for additional context. Findings appear on the Conversation and Files changed tabs, with an AI label that distinguishes them from CodeQL alerts.

  • Coverage includes gaps such as PHP, Shell/Bash, Terraform HCL, Dockerfiles, JSP and Blazor.
  • Detection areas include injection, weak cryptography, access control, sensitive-data exposure, SSRF and supply-chain risks.
  • Suggested remediation and Copilot Autofix may appear with a finding, but are not guaranteed.

Why this matters for QA engineers

Test repositories often contain shell scripts, Dockerfiles, infrastructure code and framework-specific helpers alongside application tests. The preview can expose risks in those areas earlier, but GitHub explicitly warns that AI findings may include false positives.

  • Reproduce injection and access-control findings with a focused negative test before accepting a fix.
  • Review any Autofix diff for changed test behavior, permissions, secrets handling and dependency pinning.
  • Track confirmed findings, dismissed false positives and regressions separately during the preview.

Preview limits and access

The findings are advisory and cannot currently block merges through rulesets. They apply only to pull requests, not full-repository backlog scans. GitHub says the preview requires GitHub Code Security, a Copilot license, CodeQL default setup and explicit enterprise and organization enablement. Runs consume AI credits during the preview.

Sources