GitHub Agentic Autofix and AI Scan are generally available: what they do, who gets them and how they use AI Credits

On October 15, 2026, GitHub made two AI security features generally available. Agentic autofix hands a code scanning alert to a GitHub Copilot agent, which writes a fix, reruns CodeQL to check it and opens a draft pull request. AI Scan for pull requests uses AI to flag vulnerabilities in code that GitHub's CodeQL engine doesn't analyze, such as PHP, shell scripts, Terraform and Dockerfiles. Both entered public preview in July. Both are now billed through Code Security AI Credits, and according to GitHub's docs, neither requires a Copilot license.

This story is based on GitHub's changelog posts and documentation, checked October 16, 2026.

The short version

• Agentic autofix: assign an alert, or up to 25 at once, to Copilot. It explores the codebase, fixes the issue, reruns CodeQL and opens a draft PR.
• AI Scan: runs on pull requests and adds advisory findings, labeled "AI," in languages and frameworks CodeQL doesn't cover.
• Who gets them: GitHub Code Security or GitHub Advanced Security customers, plus public repositories, with some exceptions.
• Cost: both consume Code Security AI Credits, including on public repos. Agentic autofix also uses Actions minutes. GitHub hasn't published a per-fix or per-scan rate.

What does agentic autofix do?

Classic Copilot Autofix, launched in 2024, suggests one change for one alert, and it stays free. The agentic version goes further. Per GitHub's changelog, it:

1. explores relevant files across the codebase, not just the flagged file
2. proposes a fix
3. reruns CodeQL to confirm the alert is closed, and iterates if it isn't
4. opens a draft pull request explaining the fix, why it closes the alert and what validation it ran

You can ask for changes by commenting on the PR. During the preview, GitHub said a fix "typically" took two to four minutes.

Ways to start it:

• On an alert, the free Generate Fix button is replaced by Assign to Copilot.
• Select up to 25 alerts in the security alerts list to fix together in one PR.
• Through the REST API, set assignees to ["copilot-swe-agent[bot]"] on the Update a Code Scanning Alert endpoint.

Limits. Only alerts on the repository's default branch can be assigned. GitHub's docs say agentic autofix works "on a best-effort basis." It validates with CodeQL's standard query suite, so it can't confirm fixes for alerts from custom queries, the security-extended suite or third-party tools.

What does AI Scan do?

AI Scan fills gaps in CodeQL's language coverage. According to GitHub's docs, it:

• runs when a pull request is opened and after each new commit, if code scanning and the repository's AI Scan setting are on
• covers languages and frameworks CodeQL doesn't, including PHP, Shell/Bash, Terraform (HCL), Dockerfiles, JSP for Java and Blazor for C#
• looks for issues such as injection, weak cryptography, broken access control and supply chain risks
• needs no build, and doesn't read custom instruction files such as CLAUDE.md
• shows findings next to CodeQL results, labeled "AI," usually with a suggested fix

Limits. Findings are advisory. They "do not fail the Code scanning check" and can't be used in rulesets to block merges. AI Scan doesn't run on pull requests from forks or ones Dependabot creates, doesn't scan whole repositories, and "may include false positives." Since September 16, it no longer requires CodeQL default setup.

Which plans and repositories qualify?

Last verified: October 16, 2026 (GitHub changelog, October 15, 2026; GitHub Docs on autofix, AI Scan and Advanced Security billing)

Two wording differences: the agentic autofix changelog says it's for customers "on GitHub Enterprise Cloud," while the docs also cover personal public repos. And the changelog says access "requires a Copilot billing method to be defined," while the docs say "you don't need a Copilot license" and don't mention a billing method. A billing method isn't a license, so both may hold, but GitHub doesn't spell it out. Both wordings were unchanged when rechecked on October 16.

How much do they cost in AI Credits?

GitHub doesn't publish a fixed price per fix or per scan. According to its Advanced Security billing docs:

• AI Scan and agentic autofix use the Code Security AI Credits SKU and "consume Code Security AI Credits even when the repository is public."
• If the billed account has included credits from a Copilot entitlement, those are used first. After that, usage is metered under the account's budgets.
• Usage appears under "AI usage" in enterprise billing, and admins can set budgets for the SKU, per the changelogs. GitHub's AI Scan docs add that if a hard budget leaves no credits available, AI Scan "cannot produce findings."

The billing page doesn't list a dollar rate for Code Security AI Credits specifically. For Copilot, GitHub prices one AI Credit at $0.01, as our Copilot computer use story explains. GitHub advises admins to set budgets before enabling these features widely, especially AI Scan, which runs on every PR and new commit.

How do you turn them on?

Agentic autofix, per the changelog:

1. An enterprise administrator enables the "Agentic autofix" policy in Advanced Security policies.
2. An organization or repository administrator turns it on in Code Security settings. Once an organization enables it, it's on by default for eligible repos, and repo admins can opt out.
3. A Copilot billing method must be defined, per the changelog (the docs don't mention this).

AI Scan, per the docs:

1. Enterprise owners must allow it. The enterprise policy is "Not allowed" by default, and allowing it doesn't switch it on.
2. Organization owners or security managers enable it for all eligible repos with code scanning. Repo admins can opt out, but can't opt in if the organization has it off.
3. For a personal public repo, use the AI Scan for pull requests toggle under "Code scanning" in the repo's settings.
4. The ai-scan setting can also be managed through the REST API.

Why it matters

AI tools are finding vulnerabilities faster than teams can fix them; Microsoft's October Patch Tuesday set a record of more than 1,400 CVEs (our story). GitHub's pitch is to close that loop inside the pull request.

Two cautions: AI Scan can't block a merge, so treat it as a hint, not a gate. And review agentic autofix's pull requests like any contributor's; our prompt injection explainer covers why agents that read untrusted content need guardrails.

What's confirmed, reported and uncertain

• Confirmed (GitHub): October 15 general availability, how each feature works, languages, limits, eligibility, enablement and the AI Credits billing model.
• Not stated by GitHub: a dollar rate for Code Security AI Credits, typical credits per scan or per fix, the models used, and how AI Scan findings compare with CodeQL's on false positives.
• Uncertain: agentic autofix eligibility outside Enterprise Cloud, and whether GitHub Enterprise Server will get either feature.