By ByteBot
GitHub shipped two security features to general availability on October 15: Agentic Autofix for code scanning alerts and AI Scan for pull requests. Agentic Autofix is the more significant of the two — instead of suggesting a single-line patch, Copilot now explores your entire codebase, writes a fix, reruns CodeQL to confirm the alert is closed, and opens a draft pull request with documentation. Teams can batch up to 25 alerts into one PR. Eighty-two percent of organizations are currently carrying security debt, according to Veracode’s 2026 State of Software Security Report. GitHub is betting agentic AI can close that gap at scale.
How Agentic Autofix Works
When you assign a CodeQL alert to Copilot — individually or in batches of up to 25 — the agent starts by reading the relevant files across your repository. It doesn’t just look at the flagged line. It reasons about the surrounding code, then proposes a fix, reruns CodeQL to validate the patch actually closes the alert, and iterates if it doesn’t. The result is a draft pull request with the code changes and a written explanation of what was done and why.
That’s a meaningful step up from one-line suggestions that frequently miss the same bug pattern appearing in three other files. Moreover, the batch feature compounds this: security teams can group related alerts by component or owner, assign all 25, and review a single consolidated PR rather than triaging each alert manually. GitHub’s internal cohort data suggests repositories using Code Quality features show 50% smaller backlogs within six months.
Related: GitHub Copilot HydraFusion Max: What Changed October 14
The Catch You Need to Know
Both features come with restrictions that most coverage glosses over. Agentic Autofix is available exclusively on GitHub Enterprise Cloud with a Code Security or Advanced Security license. GitHub Enterprise Server customers and anyone on GitHub Free or GitHub Pro won’t see this in their interface. Additionally, enterprise administrators must explicitly enable the “Agentic autofix” policy before organizations can access it. On top of that, both features consume AI Credits from a separate Code Security budget — not from regular Copilot credits — which means security, engineering, and finance teams need to coordinate before enabling at scale.
AI Scan carries a further limitation worth stating plainly: its findings are advisory only. They cannot block pull request merges through ruleset requirements. If your branch protection rules require passing code scanning checks, AI Scan findings don’t count. However, GitHub’s own documentation says findings “cannot yet be used in rulesets to enforce merge requirements” — the word “yet” is deliberate. Merge blocking is likely coming. For now, AI Scan raises a flag; a developer chooses whether to act on it.
GitHub AI Scan Fills the CodeQL Gap
AI Scan’s distinct value is coverage for languages CodeQL doesn’t support: PHP, Shell/Bash, Terraform HCL, Dockerfiles, JSP for Java, and Blazor for C#. It runs on pull request diffs without requiring a build system, which means it works on infrastructure-as-code repositories where a traditional build step doesn’t apply. For teams running Terraform alongside application code, this fills a real blind spot — Terraform misconfigurations and insecure Dockerfile practices have historically fallen outside CodeQL’s scope. Public repositories get AI Scan for free; private repositories need Code Security. Findings appear alongside CodeQL results with an “AI” label distinguishing them from static analysis results.
For context on why this matters: Veracode’s 2026 State of Software Security Report found that 82% of organizations carry security debt, and 60% of those have critical debt. The backlog grows faster than teams can manually remediate it. Consequently, tools that automate remediation at the codebase level — rather than flagging individual lines — represent a real change in how security debt gets addressed.
Key Takeaways
- Agentic Autofix went GA on October 15. Copilot now explores your codebase, validates fixes with a CodeQL rerun, and opens draft PRs — up to 25 alerts batched at once. This is a remediation agent, not a suggestion box.
- Enterprise-only gating. Requires GitHub Enterprise Cloud plus Code Security or Advanced Security. Not available on GitHub Free, GitHub Pro, or GitHub Enterprise Server.
- AI Scan is advisory for now. Findings cannot block merges through rulesets — yet. GitHub’s docs say “cannot yet,” signaling this will change. Until then, AI Scan informs; it doesn’t enforce.
- CodeQL coverage expands. AI Scan adds PHP, Bash, Terraform, Dockerfiles, JSP, and Blazor. Public repos get it free; private repos need Code Security.
- Separate billing. Both features consume AI Credits from the Code Security budget, not your Copilot allocation. Plan your budget before enabling broadly.
