Sanitizes user input before inclusion in LLM prompts to mitigate prompt
injection attacks. Filters common injection patterns such as instruction
overrides, system prompt markers, and role-switching attempts.
Distinct from sanitize_text_input() which targets XSS.
Usage
sanitize_llm_input(text, max_length = 2000)
Arguments
- text
Character string of user input destined for an LLM prompt
- max_length
Maximum allowed character length (default: 2000)
Value
Sanitized character string
NIST Compliance
Implements NIST SI-10 (Information Input Validation) for AI/LLM contexts.
