Check Whether Turning On GitHub AI Secret Detection in Push Protection Costs Extra

What does the new model look for?

According to the GitHub announcement, this is a model GitHub tuned for secret detection. It reads the surrounding code to find values that are likely credentials, and it does not generate code or text. A notable point is that it also targets passwords that have no recognizable format, as tokens do.

Values with no fixed format are easy to miss with pattern search. So you can understand it as a context-reading model added for those cases.

Which features are free and which use credits?

The announcement splits on the three points below. The date checked is 2026-10-11.

Source: GitHub Changelog, published 2026-10-07, checked 2026-10-11.

Per the announcement, credit usage starts within the next few weeks. The exact start date is not given. It also says billing starts only after you join the preview and turn the feature on. It says organizations already in the push protection private preview will be charged if they do not turn it off before billing starts, so check this part first.

What changes when you turn it on in push protection?

The push protection documentation explains it as a feature that blocks secrets before they enter a repository. Besides command-line pushes, commits from the web interface, file uploads, and REST API requests are also scanned. With write access, you can bypass it by giving a reason, and the bypass is recorded in alerts and the audit log.

With AI detection added, credentials with no fixed format can also be caught at push time. The announcement says this check can use credits even when it does not block a push. So the more pushes a repository gets, the more you should estimate usage in advance.

• Usage is billed to the organization that owns the repository, not deducted from an individual user's credits.
• The exception is repositories in the personal namespace of an enterprise managed user (EMU). In that case the pushing user's credits are used.
• Usage appears under the Secret Protection AI Credits item in AI usage insights.

How is Copilot /security-review different?

/security-review is a command that reviews your changes before you commit or push, in a Copilot CLI or Copilot app session. It only reads and does not modify files. A check by the secret classification model will be added to it soon.

This check is off by default and running the command does not turn it on. The announcement also says an agent must not turn on features that use credits, or change policies and budgets, without human approval. Teams that put this command into automation would do well to turn that sentence into an operating rule. That is a suggestion.

You can use this check even without a Secret Protection license. Instead, usage is charged to the billing account of the active Copilot plan, and the item is reported under Secret Protection.

Which plans can my organization use it on?

Source: GitHub Changelog, published 2026-10-07, checked 2026-10-11.

Copilot Business and Enterprise can use the /security-review check, but that does not stand in for the Secret Protection license push protection needs. The announcement adds that it varies by invitation and admin policy.

What should I settle before turning it on?

Administrators can turn the new features off by policy and also set an AI credits budget. Turning a feature on does not lift these controls. However, a budget alert does not stop usage. To make usage stop at the limit, set it yourself, such as Stop usage when budget limit is reached, wherever that setting is available.

Here is an order for checking. This is also a suggestion.

1. Check whether your organization has purchased Secret Protection or Advanced Security, and which plan it is on.
2. See whether you are already in the push protection private preview, and decide whether to keep using it.
3. Create the budget limit and the stop-at-limit setting first, then turn the feature on.
4. If anything in your automation calls /security-review, write down who has the authority to turn on the secret check.

A single announcement mentions both private preview and public preview. Before applying anything, reopen the original announcement to see which stage push protection is in.