Which dating apps train AI on your data?

Tinder lists “Developing and improving new features and services, including through machine learning and other technologies, and testing them out” among its uses. Machine learning is named as a means of building features. What members write and upload is never named as what it learns from. The entry was nearly filed as an adverse grade on a first pass. Reading it beside Hinge’s policy, its sister app in the same corporate group, is what showed the clause describes a method, not an input.

Hinge’s policy wording is identical, word for word, and the record reads the pair together. What separates them is the help centre. Three Hinge articles say the age estimate and audit images kept from a verification selfie may be used to “train select machine learning models that power our trust and safety tools”. Selfie Verification is “a voluntary process”, so that tier is B. Where Face Check or an age check is compulsory, “you must complete it to fully access Hinge” and you “cannot delete your Verification Data without deleting your account”, which is an F. Profiles, photos and messages stay open. The FAQ the policy points to says your data powers a matching algorithm, “upon which we constantly improve”, and its own page on matching says the system “learns from you and others on the app”. Neither says train.

Bumble’s policy is unusually forthcoming about mechanism. It names OpenAI under “AI tool providers to assist with message prompts”, states how long OpenAI keeps those prompts, and devotes a numbered section to “Our Use of Algorithms”, matching by profile and moderation by AI. What it never uses is the word train, in any form. So a reader learns that their message prompts reach a third party, and how long that third party keeps them, and not what either company builds from them.