GitHub Made AI Scan for Pull Requests Generally Available and It Still Will Not Fail Your Check

On October 15, 2026, GitHub’s changelog announced that Code scanning AI Scan for pull requests is generally available. It looks at PR diffs for languages and frameworks CodeQL does not cover well: PHP, Shell/Bash, Terraform HCL, Dockerfiles, JSP, Blazor, and friends. Findings show up next to CodeQL with an “AI” badge, sometimes with a risk explanation and an autofix suggestion. They are advisory. They do not fail the code scanning check. They cannot block merges through rulesets. The industry will call that “responsible AI.” Everyone who has watched a critical issue merge on a Friday will call that “decorative lint.”

This is not a hate piece about static analysis. It is a field guide to what shipping AI Scan actually means when your security theater budget already bought three other bots that also cannot say no.

What GitHub says it does

From the October 15 changelog and GitHub’s docs framing:

• Complements CodeQL by scanning PR changes in languages/frameworks CodeQL does not currently cover (PHP, Shell/Bash, Terraform/HCL, Dockerfiles, JSP for Java, Blazor for C#, “and more”)
• Runs automatically on pull requests when code scanning is on and the repo’s effective AI Scan setting is enabled
• Findings appear on Conversation and Files changed tabs with an AI indicator so you can tell them apart from CodeQL alerts
• Findings may include a security-risk explanation and a suggested autofix
• AI Scan does not run on PRs from forks or on Dependabot PRs

That last bullet matters. Fork PR supply-chain theater remains a human problem. Dependabot bump PRs do not get the AI sidekick either. Plan accordingly.

The part the sales deck will whisper past

GitHub is explicit: AI Scan findings are advisory. They do not fail the code scanning check. They cannot block pull request merges through ruleset requirements.

So you get:

1. A Finding
2. A Badge that says AI
3. Maybe a Suggested Fix
4. A Merge button that still works if CodeQL is green and humans are tired

If your control objective was “unreviewed risky PHP cannot land,” AI Scan alone does not implement that control. It implements a comment. Useful comments. Still comments.

Availability and the license tax

• Public repos (org or personal): eligible
• Private / internal org repos: need GitHub Code Security or GitHub Advanced Security
• Copilot license: not required for AI Scan

Billing: AI Scan consumes Code Security AI Credits. Usage shows under enterprise billing/licensing as the Advanced Security product’s “AI usage” section. You can set SKU-level budgets for those credits. In other words, the bot can also surprise Finance. Set a budget before the pilot becomes an invoice genre.

How to enable it without cosplaying competence

Exact click-paths move around GitHub UI releases, so treat this as a checklist against current docs, not a screenshot bible:

1. Turn on code scanning for the repository (or org default) the boring way you already should have.
2. Enable the repository’s AI Scan setting (effective setting may inherit from org policy). Confirm in docs for your plan tier.
3. Open a PR that touches an AI-Scan-covered language and verify findings appear with the AI indicator.
4. Decide, as a policy, what humans must do when AI Scan yells. Because the platform will not force the issue.
5. Watch AI Credit burn for a week. Cap it if your org treats “unlimited AI” as a personality trait.

If you only enable it on public OSS and leave private cash-cow repos dark because GHAS paperwork is stuck in procurement, congratulations: you secured the README and left the product alone.

Cynical but accurate: AI security theater vs real review

The industry loves tools that produce findings without producing friction. Friction is where bugs die. Advisory-only AI on the languages CodeQL skipped is a reasonable product bet. Marketing it as “AI secured your PR pipeline” is how you get board slides that outrun reality.

Real review still needs:

• Humans who understand the threat model of the change
• Blocking checks for high-confidence, high-impact defect classes where you have them
• Secrets scanning, dependency review, and least-privilege CI identities that are not LARPing
• A culture where “the bot said LGTM” is not a merge rationale

Autofix suggestions are accelerants. They are also a great way to land plausible-looking wrong patches if nobody reads the diff. Require the same scrutiny you would give a junior engineer’s drive-by fix.

Where this sits next to other October GitHub security chrome

GitHub also shipped related Advanced Security / code scanning metrics work around the same window (for example, AI Scan metrics in security overview called out in the October changelog stream). Metrics without merge policy are dashboards. Dashboards without owners are wallpaper. If you enable AI Scan, also decide who looks at the overview weekly and what threshold triggers a process change.

What AI Scan is good for anyway

• Catching foot-guns in Terraform and Dockerfiles that CodeQL was never going to own
• Giving PHP and shell PR authors a second set of eyes that shows up in the Files changed tab
• Creating a paper trail of “we had a finding and ignored it,” which is unpleasant and useful in post-incidents

What it is not: a replacement for CodeQL, a fork-PR shield, a Dependabot babysitter, or a ruleset enforcer.

Sloth take

GitHub shipped an AI reviewer that can spot issues in the languages the serious analyzer skipped, then politely refused to stop the merge. That is honest product design and also peak 2026 security culture: more signals, same Friday merge energy. Turn it on. Read the findings. Set a credit budget. Keep a human and a blocking check in the loop for anything that can page you at 2 a.m. The badge says AI. The risk still says yours.

Want a PR security gate that is policy, not vibes? Talk to Dark Edge Consulting.

Sources

• GitHub Changelog: AI Scan for pull requests generally available (Oct 15, 2026) – https://github.blog/changelog/2026-10-15-ai-scan-for-pull-requests-generally-available/
• GitHub Docs: AI-powered security detections / AI Scan concepts (linked from the changelog)