GitHub moves AI secret detection to a fine-tuned model and AI Credits

GitHub announced on October 7, 2026, a fine-tuned model built for detecting leaked secrets in code. The model reads surrounding code to flag likely credentials, including passwords without a recognizable token format, and it does not generate code or prose. GitHub is using it in three places: secret scanning alerts, push protection and the GitHub Copilot /security-review command, each with a different availability stage and billing rule.

What is new in GitHub's secret detection

The GitHub changelog entry (opens in a new tab) splits the rollout into three parts.

AI-detected secret alerts moved first. Customers with AI-detected Password alerts were upgraded to the new model automatically. These scans stay included in GitHub Secret Protection (GHSP) and GitHub Advanced Security (GHAS) at no additional charge.

AI-detected secrets in push protection is in private preview. Push protection checks for unstructured credentials at push time, so a developer can remove a secret before it enters repository history.

Secret checks in the Copilot /security-review command are listed as coming soon in private preview, for Copilot CLI and the Copilot app.

The billing change

The two new checks consume GitHub AI Credits. GitHub published the planned billing model ahead of broader availability and says AI Credit usage for these opt-in checks will be introduced in the coming weeks.

For push protection, credits are billed to the organization that owns the repository. A check can consume credits even when it does not block a push. The exception is user-namespace repositories for enterprise-managed users (EMUs), where usage goes against the pusher's allocated credits. Usage appears under the Secret Protection AI Credits SKU in AI usage insights.

For /security-review, the new checks do not need a GHSP or GHAS license. Usage goes to the billing account for the active Copilot plan, reported under GHSP. Billing for both starts once an organization or user opts into the public preview and enables the feature.

Who it is for and what changes in practice

Teams already paying for GHSP or GHAS get a model swap on existing AI-detected alerts without a price change. AI push protection needs GitHub Enterprise Cloud or GitHub Team with a paid GHSP or GHAS purchase, and an administrator must enable it.

Developers on individual plans of GitHub Copilot, including Pro, Pro+, Max, Free and Student, are listed as eligible for the security review checks. Copilot Business and Copilot Enterprise are eligible on supported platforms, subject to administrator policies. A Copilot subscription does not replace the GHSP license that AI push protection requires.

GitHub Enterprise Server is part of this too. The model brings AI-detected alerts to GHES 3.23 in public preview, included with existing GHSP and GHAS purchases. AI push protection and the Copilot security review command are not part of that GHES release.

How to try it

For alerts on github.com, the docs on enabling AI-detected secrets (opens in a new tab) give the path. In a repository, open Settings, then Advanced Security, and under "Secret Protection" click Enable next to "Scan for AI-detected secrets". At organization level, the same option sits in a custom security configuration. Results land in a separate list on the secret scanning alerts page.

The /security-review command runs inside a supported Copilot CLI or Copilot app session. GitHub's Copilot CLI agent docs (opens in a new tab) describe security-review as a built-in, read-only agent that reports high-confidence findings. The new secret classifier checks will be off by default, and running /security-review does not turn them on.

Spending controls are announced alongside the billing model, which GitHub says arrives in the coming weeks. According to the changelog, administrators will be able to disable the new capabilities by policy and set AI Credit budgets. For a dedicated budget, the changelog names Advanced Security as the product and Secret Protection AI Credits as the SKU. Budget alerts alone do not stop usage; the changelog points to the "Stop usage when budget limit is reached" option as the cap.

Limits and caveats

GitHub's responsible use page for security AI features (opens in a new tab) lists limits for generic secret detection. It may miss credentials checked into a repository. It may not cover test code. It may produce more false positives than partner pattern detection, so each alert needs review.

Customers in the AI push protection private preview will start consuming credits after the billing change unless they disable the feature first, according to the changelog.

What is not known yet

GitHub has not published AI Credit prices for these checks in the changelog, only the billing model. The exact start date for credit billing is given as the coming weeks.

The changelog gives no accuracy figures for the new model and no comparison with the model it replaced. The responsible use page reviewed here does not mention the fine-tuned model, so it is unclear whether its listed limits all still apply. No date is given for the /security-review secret checks entering private preview.