Glossary · Term

text masking

← all terms

Definition

Plain language

Painting over any writing in a picture before showing it to the model, so it cannot read it.

As stated in the literature

A preprocessing defense that runs text detection over an input image and occludes detected regions with solid fill, eliminating the in-scene text channel at the cost of legitimate label and signage reading.

Also called: masking, masked

Why it matters: It shuts down an entire attack channel outright, but at the price of blinding the system to legitimate signs, labels, and street names it may need to do its job.

For example, a delivery robot's camera feed could have every stretch of visible writing covered with grey boxes so that a note taped to a wall saying "go through this door" never reaches the model.

Heard on the show

“And third, the masking result is close to tautological.”
Episode 236 — Why a Printed 'OPERATOR OVERRIDE' Note Redirects Robot Planners

Mentioned in 13 episodes

  1. 236
    Why a Printed 'OPERATOR OVERRIDE' Note Redirects Robot Planners
  2. 199
    Finding a Model's Hidden Behaviors Without Knowing What You're Looking For
  3. 169
    Why Better Bug Reports Can Make AI Coding Agents Worse
  4. 167
    How Teaching an AI to Predict, Not Act, Made It a Better Actor
  5. 154
    How a 7B Model Out-Investigates a 72B One by Choosing What to Look At
  6. 151
    Why More Experience Made This AI Agent Worse, And How to Fix It
  7. 084
    Terminal Agents Get Free Supervision From The Tokens We've Been Throwing Away
  8. 073
    When Three LLMs Talk to Each Other, Their Ideas Quietly Stop Moving
  9. 068
    The OS Trick That Makes Tree Search Practical for Coding Agents
  10. 064
    When Agent Memory Stops Being a Database and Starts Being a Skill
  11. 021
    Ten Thousand Examples Beat the Full Industrial Pipeline for Search Agents
  12. 020
    The Compliance Gap: Why AI Says Yes and Does No
  13. 013
    Why Search Keeps Rediscovering the Same Workflow, and What That Means