How we use AI

Some parts of RemoveMyTracks call a language model (today: xAI for license photos, Anthropic for catalog work). This page is the honest list. Reply classification is rules, not a model. This is not legal advice.

What a model sees

• Driver's license photo — to read name, address, and date of birth into the onboarding form. You can edit every field before it is saved. If the read fails, you type the fields. We do not fill the form from on-device OCR.
• A company name you add that is not in our catalog — to look for a privacy contact. That guess is never auto-sent.
• Catalog maintenance — vendor dossiers (“who is this company?”) and replacement privacy URLs. A URL is written only after a live HTTP check. We do not mark a link fixed because the model said so.

What a model does not do

• It does not decide that a vendor confirmed deletion. That comes from their email, classified by fixed rules, or from you marking it done.
• It does not fill Ford / SayMine / OneTrust webforms. Tokenized confirmation links that arrive in our mailbox are opened by a script, not a model.
• It does not train our product on your license or your letters. API use is for running the feature, not for building a RemoveMyTracks model.
• We do not market this as dark-web scanning or as a fully automated deletion guarantee.

What you still do

If a vendor asks for a portal click, an ID upload on their site, or a signed authorization, we show you the ask and your LPOA. You finish that step. We do not upload your license to third-party portals for you.

Processors

xAI and Anthropic are listed on our subprocessors page. Questions: privacy@removemytracks.com.