Glossary · 28 Sep 2026

What is model training on your data?

← Glossary

This page covers the provider's side of the phrase: an AI provider using what you send it, such as questions, documents and answers, to improve its models. The phrase also means training a model of your own, which is a different subject. In our view, data used in training is hard to take back out, so it is worth checking the terms before sending anything sensitive.

What contracts say

Business terms from major providers commonly rule it out. Anthropic's commercial terms state: "Anthropic may not train models on Customer Content from Services". Microsoft says for its Copilot service that "Prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation LLMs".

The details matter. Anthropic's privacy centre says that if you "explicitly report feedback or bugs", for example with the thumbs up or down button, then "we may use your chats and coding sessions to train our models". Consumer products and free tiers can carry different terms from business ones.

Where the practical risk is

The free tools staff use without a contract are harder to control. One single-country survey of small firms, reported by SiteProNews with its method unpublished, found "50% of SMEs admitted to leaking company data into free public AI chatbots in order to complete work tasks". Pasting a price list into such a tool to reformat it may hand it to a provider whose terms allow training.

Training or retrieval?

Training changes the model. Retrieval doesn't. An assistant that looks up your contract to answer a question reads it at that moment. If the provider doesn't train on it, nothing stays in the model afterwards. That distinction answers most "will the AI learn our prices?" worries.

What to do

Read the training clause in every AI contract, including what happens to feedback, and check retention periods. Give staff an approved tool that covers their needs, and name the data that must never go into an unapproved one.

Quarri's security page sets out how customer data is isolated, encrypted and audited.

Sources

  1. Anthropic, "Commercial Terms of Service", effective 17 June 2025: anthropic.com
  2. Anthropic Privacy Center, "Is my data used for model training?", 18 August 2026: privacy.claude.com
  3. Microsoft Learn, "Data, Privacy, and Security for Microsoft Copilot", last updated 18 August 2026: learn.microsoft.com
  4. SiteProNews, "90% of AI-Using SMEs Worry Company Data is Training Their AI Tool", 24 August 2026 (reporting a small-business publisher's survey): sitepronews.com

Quarri is an AI-native data platform for the timber supply chain. It connects buying, production, sales and inventory for forest management, sawmill, wood products and pulp, paper and packaging operators.

See it on your own data.

Live in two weeks, on the systems you already run.