Board-friendly AI terminology.
Definitions focused on risk, control, and governance rather than deep technical details.
The input sent to an AI model. It is an outbound data transmission.
The amount of text a model can consider at once. It defines the data exposure surface.
Retrieval augmented generation. The model answers using approved documents instead of training data.
A segregated AI environment with contractual controls for logging, retention, and access.
Removal or masking of sensitive data before sending to a model.
Changes in model behavior over time that can alter outputs and risk profile.
A policy layer that inspects and blocks prompts that violate data rules.
A model that does not retain conversation history beyond the immediate request.
Evidence of prompts, outputs, and reviewer decisions for regulatory review.
A required review step before AI output can enter production systems.