GitHub announced on October 15, 2026 that Agentic Autofix for code-scanning alerts is generally available. Instead of producing a one-file suggestion, the assigned Copilot agent can explore relevant code, propose a remediation, rerun CodeQL, iterate when needed, and open a draft pull request for human review.
What changed with GitHub Agentic Autofix
When enabled, the alert view’s free Generate Fix option is replaced by Assign to Copilot. Teams can assign one default-branch code-scanning alert, or select up to 25 alerts from a repository or security campaign for one agent session and one draft PR. The PR includes the proposed change, an explanation of why it should close the alert, and the validation steps taken.
- Available to GitHub Code Security and GitHub Advanced Security customers on GitHub Enterprise Cloud, including data-residency environments.
- Requires an enterprise policy, then organization or repository enablement, plus a configured Copilot billing method.
- Consumes Code Security AI Credits and GitHub Actions minutes.
- Supports code-scanning alerts on the default branch; alerts found only on pull requests or other branches cannot be assigned.
Why CodeQL validation is evidence, not merge approval
Rerunning CodeQL is useful, reproducible evidence that the specific CodeQL alert no longer appears under the query suite used by the agent. It does not prove that a patch preserves product behavior, covers every exploit path, meets accessibility or performance expectations, or fixes findings produced by custom, security-extended, or third-party rules. GitHub’s documentation describes the workflow as best effort and says fix quality is not guaranteed for third-party-tool findings.
Why this matters for QA engineers
This shifts part of security remediation from “find a developer to investigate” to “review an evidence-bearing draft PR.” QA engineers can make that handoff safer by defining the checks an agent-created security fix must pass before merge. Start with one low-risk, well-understood default-branch alert and compare the agent’s PR against a human-authored fix before expanding usage.
A practical QA review checklist
- Read the alert and diff together: confirm every changed file is relevant and no authorization, input-validation, or error-handling behavior regressed.
- Reproduce the risk safely: use a non-production fixture or a regression test that demonstrates the vulnerable path without exposing secrets or systems.
- Run the product tests: CodeQL closure is only one signal; execute focused unit, API, UI, and integration checks that protect the affected behavior.
- Inspect the CodeQL result: capture the alert state, query suite, commit SHA, and CI output in the PR for auditability.
- Control rollout and cost: enable the policy for a pilot repository, set AI-credit budgets, and measure accepted, reverted, and escaped-fix rates.
Bottom line
GitHub Agentic Autofix can shorten the path from a CodeQL alert to a reviewable patch, but it should feed an existing quality gate rather than bypass one. Treat the draft PR and its rerun as a strong starting point; require the same behavioral and security regression evidence you would expect from any other production change.