AI Safety & Ethics

Our approach to AI

AI Waiter uses large language models (LLMs) to answer restaurant guests. We recognise the responsibility that comes with deploying AI in customer-facing roles and follow the principles below.

Principles of responsible AI

1. Transparency

Guests should always be able to tell they are talking to an AI assistant rather than a human. We encourage restaurant owners to present the assistant honestly as an automation tool.

2. Limited scope

The assistant is deliberately limited to hospitality topics: the menu, orders, reservations, opening hours and related service questions. It does not give medical, legal or financial advice and declines requests outside its purpose. Nutrition and allergen information is presented as provided by the restaurant and should be verified by guests with serious allergies directly with staff.

3. Protection from harmful content

• Platform-level prompts are designed to avoid offensive, discriminatory or dangerous responses.
• Where a restaurant sells alcohol, an age-verification gate can block the conversation until the guest confirms their age, and the assistant refuses alcohol-related requests until then.
• Restaurant owners may add their own restrictions through system prompts; prompts that direct the AI toward harmful behaviour violate the Acceptable Use Policy.

4. Privacy of conversations

Guest conversations are processed in line with our Privacy Policy. Under the API terms of our AI providers, conversation data is not used to train their models. We do not train our own models on customer conversations.

5. Human oversight

AI assists people — it does not replace them. Restaurant staff have full access to conversations, can take over any chat in real time through the live-operator console, and guests can request a phone callback from a human at any point.

AI providers

The default platform provider is Anthropic (Claude models), subject to Anthropic's privacy policy and usage policies. Restaurants may optionally connect their own key for another supported provider; in that case the chosen provider's terms apply.

Feedback

If you believe the assistant produced an unsafe or unethical response, please tell us: [email protected]

Last updated: 12 June 2026