What AI vendors do with your data

OpenAI, Anthropic, Google (the paid Gemini API), Microsoft, AWS Bedrock, xAI, Perplexity, Together AI, Fireworks AI, GitHub (Copilot Business and Enterprise) and Snowflake say on their own pages that they do not train on API or business data by default. Mistral AI and Cohere let API customers opt out in settings. The consumer products are different: ChatGPT, Claude, the free Gemini API and AI Studio, Mistral Vibe, Grok, DeepSeek, Copilot Free and Pro, and Cursor without Privacy Mode may train on what you send unless you opt out.

Every answer quotes the clause it rests on, word for word, with a link to the page. Every day the site reads each page again, hashes its readable text and, when the hash changes, looks for each quoted clause. A clause that is no longer there is flagged on the answer until a person re-reads the page.

Each watched page is read every day and reduced to its text. When the text changes the site stores the new version and shows the added and removed paragraphs. Older changes come from Internet Archive captures of the same page, dated by the capture that first shows them.

No. It is a quick way to find the clause that answers a common question and to see when it changed. Your contract, order form or data processing addendum decides, and it can differ from the public page.

On Compliance (compliance.fru.dev), the sister site: which certifications and attestations each vendor holds, with evidence. Terms covers what the vendor says it does with your data.