Cyber Safety

GitHub Adds On-Demand Secret Scanning for AI Coding Work

A programmer noticing a security alert while reviewing a code change on a laptop, bright office
A programmer noticing a security alert while reviewing a code change on a laptop, bright office

GitHub added on-demand secret scanning to Copilot's security-review workflow in a November 17 public preview. The feature runs GitHub's secret detectors alongside the existing model-based review. It is intended to catch exposed credentials in proposed code changes before those changes are merged.

What the preview checks

GitHub says the review examines changes rather than searching the repository's full history. The feature is off by default and requires an opt-in on supported surfaces, including Copilot CLI, app sessions, SDK integrations and code review. Scans use AI credits under the applicable billing arrangement.

Those boundaries matter. Turning on a change review should not be treated as proof that an old repository contains no leaked secrets.

Why AI-generated code still needs credential checks

An assistant may copy an example, produce a configuration file or include diagnostic output in a patch. Any of those can accidentally expose a token or password. A dedicated detector can add a useful layer, but false negatives and unsupported secret formats remain possible.

The correct response to a real exposure is not simply deleting the visible line. A credential may already have been copied or used, so the owner needs to follow the provider's revocation and replacement process.

How to add the check sensibly

Review the current opt-in instructions and billing before enabling the preview. Test it with a safe, nonfunctional example and verify how findings reach the reviewer. Keep existing secret-management and repository-protection controls in place.

For teams using agents, make secret review part of the normal change process rather than an occasional cleanup. The update is useful because it brings a specialized detector closer to where code is created. It complements careful credential handling and human review; it does not make either unnecessary.

Sources and related reading

Source: GitHub Changelog, published 17 November 2026.

Related on TechnoVibes:

More coverage: Cyber Safety.

Image: AI-generated editorial illustration. It depicts a representative scene, not a photograph of the reported event or an exact product reference.