insecure-output-handling

To prevent XSS when rendering LLM outputs, treat all generated content as untrusted input by enforcing DOM-safe sinks and applying HTML sanitization before injecting any LLM-derived text into the UI.

Second-order injection in LLM pipelines occurs when generated outputs are passed unsanitized to downstream APIs or databases, prevented by applying parameterized SQL usage and strict allowlists for shell and database interactions.

Sanitizing LLM responses for code execution requires applying universal sanitation rules to the output before consumption, ensuring parameterized queries and shell allowlists are used to meet security best practices.

Yes, you need parameterized queries for LLM-generated SQL in dashboards to prevent command and second-order injection, treating LLM output as untrusted input to ensure safe downstream database interactions.

The best way to secure LLM outputs is treating them as untrusted input across all surfaces, implementing HTML sanitization for UI rendering, safe DOM sinks, and parameterized queries for automation pipelines.

Treating LLM output as untrusted input prevents command injection by forcing the application to apply universal sanitation rules and shell allowlists before executing any generated commands in downstream integrations.