GitHub Code Scanning Now Flags AI App Risks Like Prompt Injection

On 28 October 2026 GitHub announced that code scanning can now detect vulnerabilities specific to AI applications and agents. The feature is in public preview, and scans run automatically on pull requests alongside existing CodeQL analysis.

Covered risks include indirect prompt injection (XPIA) — the top risk in the OWASP Top 10 for LLM Applications, where malicious instructions hide in untrusted content such as a file, webpage or issue comment; sensitive information disclosure, where an agent can reach data without proper authorization controls; and excessive agency, where an agent has more permissions or autonomy than its task needs.

The detector flags code paths where untrusted content can reach a model with access to tools, data or business systems, and shows mitigation guidance directly in the PR. It requires a GitHub Advanced Security license and isn't available on GitHub Enterprise Server yet.

Source: GitHub Changelog.