Canary Tokens

Intermediate

Detecting unauthorized model outputs or data leaks.

Full Definition

Detecting unauthorized model outputs or data leaks.

Keywords

leakage detection

Domains

AI Economics & Strategy

Related Terms

Compute Scaling related to

Increasing model capacity via compute.

 Data Scaling related to

Increasing performance via more data.

 Chinchilla Scaling related to

Scaling law optimizing compute vs data.

 Inference Cost related to

Cost to run models in production.

 Model Inversion related to

Inferring sensitive features of training data.

 Prompt Leakage related to

Extracting system prompts or hidden instructions.

 Supply Chain Attack related to

Compromising AI systems via libraries, models, or datasets.

 Model Watermarking related to

Embedding signals to prove model ownership.

Concept Map

See how Canary Tokens connects to other concepts.

Open Knowledge Graph