Glossary · Term

black-box access

← all terms

Definition

Plain language

Being able to send something questions and see its answers, without seeing anything about how it works inside.

As stated in the literature

An attacker or evaluator model in which only sampled outputs are observable — no weights, activations, or token probabilities. Contrasts with white-box access.

Also called: black-box, black box access

Why it matters: It defines the realistic threat model for most deployed AI systems, since anyone with an ordinary account can mount attacks that need nothing more than questions and answers.

For example, someone using a chatbot through a public website can type prompts and read replies, but has no way to inspect the model's internal numbers.

Heard on the show

“There's one recent image-only line in visual-document retrieval, but its main attack leans on gradient-based optimization, and that degrades badly under black-box transfer.”
Episode 247 — One Edited Photo, an Honest Caption, and a RAG System That Believes It

Mentioned in 14 episodes

  1. 247
    One Edited Photo, an Honest Caption, and a RAG System That Believes It
  2. 246
    160 Perfect Refusals, And The Refusals Were The Leak
  3. 229
    One Word Flips a Chatbot From Backbone to Yes-Man
  4. 226
    How a Speed Feature Lets a Stranger Poison Your AI's Answer
  5. 186
    How a Frozen Model Went From 2% to 77% on Physics Puzzles — Without Retraining
  6. 168
    When Turning Experience Into Code Makes Your AI Agent Dumber
  7. 161
    A Robot That Plays Before You Give It a Job, And Why That Beats Retrying
  8. 152
    Training a Model to Mean What It Says, And Why That Isn't the Same as Being Good
  9. 146
    How an Innocent README Can Freeze an AI Agent's Safety Check for an Hour
  10. 119
    Beating Reinforcement Learning Without Ever Touching the Model's Weights
  11. 098
    Finding Millions of Readable Concepts Inside a Real, Deployed AI Model
  12. 090
    How MiniMax-M2 Bets That Sparsity Plus Verifiable Rewards Can Match Frontier Agents
  13. 073
    When Three LLMs Talk to Each Other, Their Ideas Quietly Stop Moving
  14. 065
    One Loop to Optimize Them All: A Universal API for LLM-Driven Discovery

Related concepts

Related terms