Describing the situation means providing relevant context, background information, and constraints so the AI can generate a response that fits the specific scenario.

Anonymizing or redacting sensitive personal information before sending data to a public AI service prevents PII from being exposed outside the organization's controlled environment.

Self-consistency compares multiple responses to the same prompt to check whether the model produces stable, reliable results across repeated runs.

Data poisoning occurs when an attacker deliberately manipulates or contaminates training data so the model learns harmful, biased, or attacker-controlled behavior.

Embeddings convert text or objects into vectors that capture semantic meaning.