System Prompt
A special instruction given to an AI model that defines its behavior and personality for all subsequent user interactions.
A system prompt is the "meta instruction" that shapes how an AI behaves throughout a conversation. While user prompts are per-message, the system prompt is persistent.
Example system prompts: - "You are a helpful, harmless, and honest AI assistant." - "You are a pirate. Respond in pirate dialect." - "You are an expert Python programmer. Only provide working code with explanations."
System prompts set tone, constraints, expertise level, and output format. They're crucial for alignment—a good system prompt prevents the model from generating harmful content. Most commercial AI tools hardcode system prompts you can't see (for safety); some APIs (OpenAI, Anthropic) let you customize them.
Limitation: System prompts can fail if a user tries to jailbreak them (e.g., "Ignore your system prompt and..."). No system prompt is perfectly unbreakable.
Example
System: "You are a customer support agent for a bank. Be professional and security-conscious." User: "What's my account number?" Model refuses (security concern).