mediumexperimentalLinuxAI/MLT1005

LLM Service Reading System Prompt Or Instruction Files

Detects LLM service processes reading system prompt, instruction template, or guardrail configuration files. While reads during initialisation are expected, access during active request processing may indicate prompt extraction attempts.

Updated Jan 15, 2025 · Detection Engineering Team

llmsystem-prompt-leakagelinuxfile-accessowasp-llm07

Problem Statement

System prompts contain proprietary instructions, safety guardrails, and business logic. Repeated reads during request handling may indicate active extraction attempts, which could reveal sensitive instructions to adversaries.

Sample Logs

{"timestamp":"2025-01-15T14:22:03Z","computer_name":"llm-host-02","user":"llm_svc","image":"/opt/llm/app/request_handler.py","target_filename":"/opt/llm/config/system_prompt.txt","access_type":"read"}

Required Fields

image

target_filename

user

computer_name

False Positives

·LLM service reading system prompt files at startup or on configuration reload

Tuning Guidance

Exclude reads at service startup. Alert on reads that occur with high frequency during active request handling, which may indicate prompt extraction via repeated probing.