Legal
AI Use Policy
How Pinevon and Anchor use AI: what is sent to providers, what we do not do, the limits of the answers and how to reach a human.
Version 0.2-draft · No effective date set yet · All legal documents
Draft — pending legal review. This document is a complete working draft prepared by the Pinevon team. It has not yet been reviewed by a lawyer and no effective date has been set. It describes how the service is actually built and operated today; bracket-free wording is deliberate, but jurisdiction-specific terms must be confirmed by qualified counsel before it is relied on.
1. Where AI is used
- The Pinevon Assistant on pinevon.com answers questions about Pinevon and its products using public information.
- In Anchor, an assistant can draft replies to a business’s customers using that business’s own catalog and knowledge, and help staff prepare quotations and follow-ups.
2. What is sent to AI providers
To generate a reply we send the conversation and the relevant knowledge passages to an AI provider (see Subprocessors). We do not send unrelated account data, passwords or payment details. Provider keys are kept on our servers and never in your browser.
3. What we do not do
- We do not use customer content to train general-purpose AI models, and our provider agreements prohibit the provider from doing so with API inputs.
- We do not let one business’s data be retrieved for another business; access is checked before knowledge is retrieved.
- We do not let the AI complete a privileged action (for example change prices or send a quotation) without the permission and confirmation that action requires.
- We do not present AI output as if it came from a person, and we do not use it for automated decisions with legal effect.
4. Limits and accuracy
AI can be wrong, out of date or incomplete. Answers should say when they are uncertain and cite the source where available. Do not rely on them for legal, financial, medical or safety decisions. Prices, availability and account state shown by a person or the system of record override anything an AI says.
5. Safety controls
- Untrusted input (user messages, imported documents, web content) is treated as data, not instructions, to reduce prompt-injection risk.
- Requests are size-limited, rate-limited and cost-bounded; providers fail over safely.
- Outputs are sanitized before display.
- Usage, latency, errors and cost are measured; quality problems are reviewed.
6. Human help
You can ask for a person at any time, from the assistant or at support@pinevon.com. Report a harmful or incorrect answer to the same address with the time and text.
7. Your responsibilities
Do not put secrets, payment card numbers or special-category data into the assistant. If you use AI features to message your own customers, you are responsible for lawful, accurate communication and for any consent required by the channel or by law.