Glossary · Term

system prompt

← all terms

Definition

Plain language

Hidden instructions placed at the start of a conversation that shape how the model behaves.

As stated in the literature

A prompt segment, often invisible to end users, that sets persistent context, persona, or behavioral constraints for an LLM session.

Also called: system prompts

Why it matters: It is the main lever developers have for shaping model behavior in deployment, and a major target for prompt-injection attacks.

For example, a customer-service bot might have a hidden system prompt telling it 'You are an airline agent; never discuss competitors.'

Heard on the show

“A hundred and sixty times, researchers asked eight frontier models to hand over the secret PIN sitting in their system prompt.”
Episode 246 — 160 Perfect Refusals, And The Refusals Were The Leak

Mentioned in 30 episodes

  1. 246
    160 Perfect Refusals, And The Refusals Were The Leak
  2. 240
    Frontier Models Designed Follow-Ups To Fraudulent Papers 93% Of The Time
  3. 236
    Why a Printed 'OPERATOR OVERRIDE' Note Redirects Robot Planners
  4. 235
    Why Chatbot Safety Erodes 350 Messages Into a Real Conversation
  5. 228
    Same Chatbot, Two Doors: Why 'Grok's Opinion' Doesn't Exist
  6. 221
    Two Hundred Clean Economics Answers, And a Model That Endorses Race Science
  7. 219
    Forty-Four AI Models, One Word, And The Newest Ones Conform Most
  8. 208
    The Blank Space in Your AI Approval Box That Isn't Empty
  9. 185
    Aligned to Refuse, Built to Tap: When Phone Agents Know the Task Is a Crime and Do It Anyway
  10. 157
    When an AI Coding Agent Drives a Phone Through the Terminal, No Screen Needed
  11. 146
    How an Innocent README Can Freeze an AI Agent's Safety Check for an Hour
  12. 123
    Five Identical Worlds, One Swapped Model: What Happens When AI Agents Run for Fifteen Days
  13. 121
    When the Agent Says It's Done But Nothing Happened: Debugging the Harness, Not the Model
  14. 120
    How an AI Agent Rewrites Its Own Tools, Without an Answer Key
  15. 107
    How a Market of Crippled AI Agents Outscored One Unrestricted Model
  16. 096
    How Treating an AI Agent's Execution Like Git Recovers a Coordination Penalty
  17. 090
    How MiniMax-M2 Bets That Sparsity Plus Verifiable Rewards Can Match Frontier Agents
  18. 078
    Training a Markdown File: When LLM Self-Improvement Borrows the Discipline of Neural Net Training
  19. 060
    When Splitting One Model Across Three Agents Doubles Its Accuracy
  20. 047
    When Agent Benchmarks Lie: The Harness Problem in Open-Source AI
  21. 046
    When the AI Optimizer Edits the Grade Book: Why Harnessing Evolution Needs a Wall
  22. 045
    When a Frontier Model Talks Its Own Twin Into Climate Denial
  23. 044
    How One Sentence and a Forged History Flip the Most Aligned Models
  24. 039
    When Smarter Agents Get Fooled by Three Extra Nodes in a Database
  25. 034
    Catching Multi-Agent Deadlocks Before Deployment With a 40-Year-Old Tool
  26. 030
    Why Your AI Agent Won't Stop Working — and Each Model Falls for a Different Trap
  27. 022
    Training the Model Spec Directly: An Alignment Lever Aimed at the Say-Do Gap
  28. 017
    When the Agent Grades Its Own Homework: A Brutal New Benchmark for AI Workers
  29. 009
    How Two Silent Library Bugs Quietly Invalidated a Wave of Reasoning Papers
  30. 001
    When AI Models Quietly Protect Each Other From Shutdown