Which gaming platforms train AI on your data?
Of the 5 gaming platforms on file, 2 are graded, and both of them train on your data by default. The other 3 have a policy that doesn't say either way. The best graded is Epic Games (D) and the worst is Roblox (F).
Do they train on your chat, your voice and how you play?
Every story about gaming privacy is about what a console or a launcher collects. The companies graded here hold what you type to strangers, what you say over voice chat and how you play. They’re graded on a narrower question: does any of that train a model, and how hard is it to stop. Read together, they divide by what the company’s own documents say. Enough to grade, which means training with no way out or training with a way out you have to write for. Or nothing that reaches the question at all.
The ones that name the training
Roblox uses comments and messages for “training and improving our filter technology” on a legitimate-interest basis, with no consent asked and no setting named. Voice is the exception: recordings inform training “with your consent”. The headline puts it in a sentence: voice needs consent, text doesn’t. On a platform whose users are largely children, the split runs the wrong way round, and no setting in the document keeps what a child types out of the training set.
Epic names the training in two purposes: “training machine learning technologies to detect and moderate illegal, harmful, toxic, and fraudulent conduct and content”, and improving “the machine learning technologies that help deliver these personalized recommendations”. Both draw on “your information”, which the policy says includes what you post in chat and forums and how you play. The way out is a rights request by email to its privacy team, not a setting. On by default, with an opt-out you have to go looking for: a D on this site.
The ones that never raise it
Nintendo’s privacy policy doesn’t use the words train, AI or machine learning once. Whether your play, voice chat or purchases train anything isn’t stated. A policy that doesn’t raise the subject can’t be read either way. This is a finding of silence, not an inference.
Valve’s Steam privacy agreement is long and never mentions training, models or AI. It does say that store content and update messages are “tailored to meet your needs and populated with relevant recommendations” from the data it describes. Recommendations aren’t training, and the document doesn’t say what else your reviews, chat and play data feed. A finding of silence.
Unity’s player policy is mostly about the advertising and analytics identifiers it collects inside other companies’ games, and advertising is not training. Its only AI language is a heading and one sentence: “We may use third-party services that utilize artificial intelligence or other advanced technologies to help us operate our business and support internal functions.” That’s Unity’s back office, not your data as a training input. “Training” appears twice, both times meaning employee training. Whether player data trains anything is not stated.