Skip to content
AdminKit

System prompt

What is System prompt?

A system prompt is the instruction set an application gives an AI model before a task starts, setting its role and behavior boundaries, separate from user input and content the model reads afterward. AdminKit's guide to securing an MCP server flags tool output that merely imitates a system prompt as a red flag.

By AdminKit team, Netbase JSC · Updated

A system prompt is the instruction set an application gives an AI model before a conversation or task starts, establishing its role, behavior boundaries and what it’s permitted to do, as distinct from the user’s own input and from any content the model reads afterward while performing a task. Anthropic’s prompting guide, for example, recommends setting the model’s role in the system prompt to focus its behavior and tone. Because a model reads its system prompt and the content it later encounters — a web page, a document, a tool’s response — through the same text channel, it doesn’t automatically know which parts are trusted instructions and which are untrusted data, which is the root cause of prompt injection, OWASP’s top-ranked risk for LLM applications and also a cataloged adversary technique in MITRE ATLAS, the ATT&CK-style knowledge base of real-world attacks against AI systems — AdminKit’s own prompt injection glossary entry covers that root cause in more depth, and it is especially relevant to MCP tool output, which a model reads through the same channel as its system prompt. A related risk, system prompt leakage, covers what happens when the prompt itself contains sensitive instructions an attacker can extract. AdminKit’s guide to securing an MCP server flags the injection risk directly: text that merely resembles system-prompt-style instructions, embedded in otherwise ordinary tool output, is a common red flag worth filtering before it reaches the model, precisely because the model can’t reliably tell a genuine system prompt from an imitation.

System prompt is part of AdminKit’s AI agents & MCP cluster; the full AdminKit glossary lists other related terms.

Frequently asked questions

What does a system prompt set?

An AI model's role and behavior boundaries before a task starts, kept separate from user input and from content the model reads afterward.

Why can a model struggle to tell its system prompt apart from untrusted content?

Because it reads both the system prompt and later content, such as a web page or tool response, through the same text channel -- the root cause of prompt injection.

What is system prompt leakage?

A related risk where the system prompt itself contains sensitive instructions that an attacker can extract from the model.

What should an MCP server treat as a prompt-injection red flag?

Text embedded in otherwise ordinary tool output that merely resembles system-prompt-style instructions.