GitHub makes agentic autofix for code scanning alerts generally available

GitHub announced on October 15 that agentic autofix for code scanning alerts is generally available. The feature hands a security alert to Copilot's cloud agent, which works through the codebase to produce and verify a fix instead of suggesting a patch in isolation.

How it works

When a developer assigns an alert to Copilot, the agent explores the relevant files, drafts a change and reruns CodeQL to check that the alert is actually closed. If the check fails, it iterates. The result is a draft pull request that summarises the fix, explains why it resolves the alert and lists the validation steps. Developers can request changes by commenting on the pull request or through the session in the Agents tab. Up to 25 alerts can be grouped into a single pull request.

During the preview, GitHub said fixes typically took two to four minutes and could be triggered for single alerts, multiple alerts, security campaigns or through the REST API. An editor's note on the preview announcement stated that the feature covers alerts from both CodeQL and third-party scanners.

Who can use it and what it costs

General availability covers GitHub Code Security and GitHub Advanced Security customers on GitHub Enterprise Cloud, including the data-residency offering. An enterprise administrator must enable the policy first, after which organisation or repository admins can switch it on, and a Copilot billing method must be configured. Usage draws on AI Credits from the Code Security budget and on GitHub Actions minutes; the changelog does not list unit prices, but administrators can set budgets for these credits.

Once enabled, an "Assign to Copilot" button replaces the free "Generate Fix" option on individual alerts, so organisations moving to the agentic flow give up the free one-shot suggestion.

Limits

The feature only handles alerts on a repository's default branch; alerts that exist only on pull requests or other branches cannot be assigned. Fixes still arrive as drafts, so human review remains part of the process. Because the agent validates against CodeQL rather than a full test suite, a closed alert does not by itself guarantee that application behaviour is unchanged, and teams will want their own CI checks to run on these pull requests.