AML.T0069

MITRE ATLAS / Discover LLM System Information

Adversaries work out details of a language model's system configuration, including its system prompt, its available tools, its guardrails, and the model version behind the interface.

Think of it likeGetting the front desk to read out its own standing instructions before you make your request.

In plain English

System prompts leak. Once the adversary knows the instructions, the tool list, and the model version, every later attempt can be written specifically against that configuration.