Glossary · Term

root

← all terms

Definition

Plain language

The all-powerful administrator account on a computer that can change anything.

As stated in the literature

The superuser account on Unix-like systems with unrestricted privileges; escalation to root is the high-value target in privilege-escalation attacks and in the agent-meltdown cascade case study.

Why it matters: Because it has unlimited control, gaining it is the prize attackers chase and the worst-case outcome when an agent's permissions spiral out of control.

For example, an attacker who gains the root account can read any file, install anything, or shut the whole machine down.

Heard on the show

“One is getting the whole root instead of snapping off the leaves so it grows back.”
Episode 232 — Coding Models Can Find the Bad Line, They Just Won't Delete It

Mentioned in 23 episodes

  1. 232
    Coding Models Can Find the Bad Line, They Just Won't Delete It
  2. 188
    A Coding Agent Found a Hole in a Peer-Reviewed STOC Proof for Five Dollars
  3. 177
    Why Raw Profiler Data Made an AI Worse at Writing GPU Code
  4. 169
    Why Better Bug Reports Can Make AI Coding Agents Worse
  5. 166
    A Router That Beats the Frontier Models It Calls
  6. 160
    Training an AI to Take Its Own Notes, So Its Future Self Works Better
  7. 148
    Why Letting an AI Watch Its Own Scoreboard Can Quietly Overwrite Its Safety
  8. 141
    How Two Tokens Reopened a Reasoning Method the Field Had Given Up On
  9. 139
    When Optimizing One GPU Kernel Quietly Breaks the Whole System
  10. 131
    Why Autonomous Research Agents Forget Their Own Lessons, and Arbor's Fix
  11. 121
    When the Agent Says It's Done But Nothing Happened: Debugging the Harness, Not the Model
  12. 117
    How an Open AI System Verified 672 Hard Math Proofs for Under $300
  13. 108
    The Reasoning Cliff: Why Thinking Longer Makes Models Worse at Exact Step-by-Step Tasks
  14. 101
    Treating Math Formalization Like a Codebase, and Where the Agents Cheat
  15. 096
    How Treating an AI Agent's Execution Like Git Recovers a Coordination Penalty
  16. 086
    Why Frozen-Weight Agents Still Get Worse Over Time
  17. 082
    Training a Deep Research Agent on 8,000 Synthetic Tasks: The Rubric Tree Trick
  18. 071
    When the Model Is Fine and the Plumbing Is Broken: Fixing Agents at the Interface
  19. 049
    An AI Agent Reached for Root in Twelve Minutes, Without Being Attacked
  20. 042
    An Agentic Scientific Computing System That Actually Remembers What It Learns
  21. 028
    Teaching a Model to Hire Copies of Itself: Recursive Agent Optimization
  22. 014
    Why a Constrained Pipeline Beat a Full Coding Agent at Finding Bugs 30-to-1
  23. 010
    When Reward Climbs But Reasoning Goes Generic: Diagnosing Template Collapse in Agentic RL

Related terms

agent meltdown privilege escalation