Source · release

Secret scanning on demand for developers and agents

Summary

Copilot's security review specialist can now run GitHub's secret scanning detectors alongside its LLM review (public preview, 2026-11-17). It checks your changes, not the repository history, for over 500 known secret formats and uses a fine-tuned model for credentials without a fixed format, such as passwords in connection strings.

Key points

  • Run it: in the Copilot CLI or app, type /security-review; agents can delegate to it through the task tool with agent_type: "security-review". In the Copilot SDK, call session.rpc.tasks.startAgent with agentType: "security-review".
  • Enable (off by default): CLI /settings aiSecretDetection on; app Settings → Experimental → AI secret detection; SDK feature flag SECURITY_REVIEW_FINE_TUNED_MODEL=true; code review: repository Settings → Copilot → Code review → Secret scanning.
  • No extra plan needed beyond Copilot security or code review; scans use AI credits under the Secret Protection AI Credit SKU.
  • Good for: a pre-commit check before an agent's branch is pushed.

Why it matters

Agents that write config and integration code often paste credentials. A detector the agent itself can call closes that gap before the push, not after the leak.