Glossary · Term

sandbox

← all terms

Definition

Plain language

An isolated computing environment where AI code can run without affecting anything important.

As stated in the literature

An isolated execution environment, often a container or VM, used to run untrusted or experimental code with controlled access to resources.

Why it matters: Sandboxes are the first line of defense whenever you let an AI execute arbitrary code, since they bound the blast radius of mistakes or attacks.

For example, when an AI agent wants to run a downloaded script, it executes it inside a sandboxed container that can't see the user's home directory or the wider network.

Heard on the show

“The thing that refused, when anything refused at all, wasn't the sandbox.”
Episode 227 — Poisoned Bug Reports Fooled Coding Agents Two Times Out of Three

Mentioned in 30 episodes

  1. 227
    Poisoned Bug Reports Fooled Coding Agents Two Times Out of Three
  2. 195
    Why 'Be Careful' Does Nothing for AI Coding Agents, and What Does
  3. 190
    The Skill Every AI Manager Is Missing: Handing Out Exactly the Right Keys
  4. 184
    An AI Built an Undetectable Secret Channel, And Another AI Couldn't Find It
  5. 182
    How a Tiny Model Too Weak to Plan Cuts a Bigger Agent's Hallucinations by 80%
  6. 168
    When Turning Experience Into Code Makes Your AI Agent Dumber
  7. 163
    Why Training Only on Perfect Solutions Cripples a Model's Reasoning
  8. 159
    Can a Coding Agent Run Its Own Robot Experiments Overnight, With No Human Resetting the Scene?
  9. 152
    Training a Model to Mean What It Says, And Why That Isn't the Same as Being Good
  10. 148
    Why Letting an AI Watch Its Own Scoreboard Can Quietly Overwrite Its Safety
  11. 139
    When Optimizing One GPU Kernel Quietly Breaks the Whole System
  12. 126
    How Coding Agents Can Mine Their Own Failures Into a Self-Targeting Curriculum
  13. 123
    Five Identical Worlds, One Swapped Model: What Happens When AI Agents Run for Fifteen Days
  14. 121
    When the Agent Says It's Done But Nothing Happened: Debugging the Harness, Not the Model
  15. 113
    What If a Prompt Injection Never Left? Attacks That Wait in Agent Memory
  16. 112
    When an AI Agent Cheats Without Being Told: Inside the Meta-Agent Challenge
  17. 111
    How a 4B Web Agent Beat Models 60x Its Size on 500 Demonstrations
  18. 105
    The Trojan Is Your Agent's Memory: Why Single-Step Defenses Miss Persistent Attacks
  19. 098
    Finding Millions of Readable Concepts Inside a Real, Deployed AI Model
  20. 097
    Same Tokens, Same Cost, Wildly Different Results: What Actually Scales in AI Agents
  21. 096
    How Treating an AI Agent's Execution Like Git Recovers a Coordination Penalty
  22. 090
    How MiniMax-M2 Bets That Sparsity Plus Verifiable Rewards Can Match Frontier Agents
  23. 076
    Same Model, Organized Differently: How an Agent Architecture Beat Frontier Systems at Research Math
  24. 068
    The OS Trick That Makes Tree Search Practical for Coding Agents
  25. 066
    Why Giving an AI Agent More Tools Can Make It Worse at Using a Computer
  26. 062
    Treating Hallucinations as Exploits: A Gate-Based Architecture for Agent Safety
  27. 061
    When Helpful Agents Go Sideways: A 404 Error, Campus Security, and Why Alignment Misses This
  28. 057
    How Uber Caught 206 Leaked Credentials With an LLM-Powered Security Stack
  29. 047
    When Agent Benchmarks Lie: The Harness Problem in Open-Source AI
  30. 013
    Why Search Keeps Rediscovering the Same Workflow, and What That Means