GitHub opened a public preview of code scanning for AI applications, flagging prompt injection in pull requests

Archive entry: this covers an event from October 28, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On October 28, 2026, GitHub opened a public preview of code scanning for AI applications on github.com. The feature searched agent code for AI-specific vulnerabilities and ran automatically on pull requests. The announcement is in the GitHub changelog entry on code scanning for AI applications.

What changed

• It detected indirect prompt injection (XPIA), sensitive information disclosure, excessive agency and other AI-specific risks.
• It flagged agent code paths where untrusted content could reach a model that had access to tools, data or business systems.
• Scans ran automatically when pull requests were created, alongside existing CodeQL analysis.
• Mitigation guidance appeared directly in the pull request.

What it meant for people building with AI

Static analysis covered indirect prompt injection, which GitHub described as the top risk in the OWASP Top 10 for LLM Applications, before code shipped. For teams building agents that read untrusted content and also reach tools, data or business systems, the check pointed at exactly those code paths during pull request review, next to the existing CodeQL results and with mitigation guidance in the same place.

Availability

Code scanning for AI applications was in public preview on github.com and was not available on GitHub Enterprise Server. It required a GitHub Advanced Security license.

Sources

• Code scanning for AI applications in public preview (GitHub, October 28, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.