AI policy
This policy describes how Northwind Assistants uses artificial intelligence in its product and what we commit to customers. It applies to every model we deploy, whether built by us or provided by a vendor.
Customer data and training
We do not train, fine-tune or evaluate models on customer conversation data without opt-in written consent. Our model providers are contractually prohibited from retaining customer prompts or using them for training; zero data retention is enabled on every provider account.
Human oversight and safety
Customers decide when the assistant may act autonomously. Consequential actions such as refunds or account changes require human approval by default. We red-team every major release for prompt injection, data leakage and harmful output, and track results in our model card.
Prohibited uses and transparency
The assistant must not be used for biometric identification, social scoring or automated decisions with legal effect without human review. Assistant responses are labelled as AI generated. We publish a transparency report each year and log model and prompt versions for audit. Report AI safety concerns to security@northwind.example.