GitHub Copilot: Secret Scanning on Demand for Developers and Agents

GitHub CopilotView original changelog

GitHub Copilot's security review specialist now runs GitHub secret scanning, so developers and agents can catch secrets in code changes before they are merged. The public preview works in Copilot CLI, the Copilot app, Copilot SDK integrations and Copilot code review, and is off by default until a user opts in. It combines 500+ provider patterns with a fine-tuned secret detection model.

Key Takeaways

  • Copilot's security review specialist now runs GitHub secret scanning on demand, covering both developer-written and agent-written changes.
  • Detection combines 500+ provider patterns from 150+ partners with a fine-tuned model that finds secrets lacking a recognizable format, such as passwords in connection strings.
  • The feature is off by default and must be enabled per tool, for example with /settings aiSecretDetection on in Copilot CLI.
  • It works in Copilot CLI, the Copilot app, Copilot SDK integrations and Copilot code review, but VS Code is not included yet.
  • The scan checks only the changes being made and does not search repository history.
  • Scans consume AI credits under the Secret Protection AI Credit SKU, with no additional plan required for the listed Copilot experiences.

What Is New

GitHub framed the update around a simple idea: secret scanning should keep pace with how software is built, whether a developer writes the code or an AI agent does. Copilot's security review specialist already looked for exposed secrets as part of its LLM-based review. Once a user opts in, it now also runs GitHub's secret scanning detectors alongside that analysis. The feature is in public preview.

Two detection methods are combined. Provider patterns detect more than 500 secret types from GitHub's catalog, built with more than 150 industry partners through the secret scanning partner program. A GitHub fine-tuned model handles credentials that have no recognizable format by reading the surrounding code, which helps find passwords and credentials embedded in connection strings. The review checks the changes being made and does not search repository history.

Where It Fits in a Workflow

In Copilot CLI and the Copilot app, a user can ask the coding agent to run the security-review specialist, then use its findings to make fixes and request another review. They can also run /security-review manually on their own changes or on an agent's work.

Developers building their own agents with the Copilot SDK can have the agent request a security review before it returns changes, without maintaining a separate set of secret detectors. In Copilot code review, secrets can be flagged on added lines during pull request review, including changes that were never checked during development.

How to Turn It On

Secret scanning is off by default. Until a user opts in, security review keeps using only its existing LLM-based detection.

  • Copilot CLI: in an interactive session, run /settings aiSecretDetection on, then delegate to security-review through the task tool or run /security-review.
  • Copilot app: in Settings > Experimental, turn on AI secret detection, then delegate to security-review or run /security-review.
  • Copilot SDK integrations: set the SECURITY_REVIEW_FINE_TUNED_MODEL feature flag to true when creating a session and configure GitHub authentication on the SDK client. A model-provider API key alone is not enough, and tokens from an integration's own OAuth app are not currently supported.
  • Copilot code review: in the repository's Settings > Copilot > Code review, turn on Secret scanning under Security, then request a review or set up automatic reviews.

VS Code support is not included in this release. GitHub updated the post on November 18, 2026 to clarify the SDK authentication requirements.

Availability and Billing

No additional plan is required for the listed Copilot experiences. Scans consume AI credits under the Secret Protection AI Credit SKU.

Copilot Secret Scanning on Demand for Devs and Agents | Yet Another Changelog