The policy doesn't say. That's the finding.

23andMe holds genotypes. Its privacy statement runs to 23,000 characters and does not use the words "train", "AI" or "machine learning" once.

Abridge's privacy policy never states whether your content is used to train AI models — a documented absence, not an unread document.

ADT lists faceprints, voiceprints and camera audio among what it collects. Its policy never uses "train", "AI" or "machine learning" — the training question goes unanswered.

Amazon's privacy notice runs to 28,000 characters and does not use the words "train", "AI" or "machine learning" once.

Audible knows every book you have listened to and where you stopped. Its privacy notice never mentions AI, machine learning or training.

Backblaze backs up entire hard drives. Its privacy policy contains no reference to AI, machine learning or training.

Bank of America's privacy notices never mention AI, machine learning or model training. Its Erica page says Erica doesn't use generative AI, not what trains it.

Barclays' full privacy notice says it may run your information through AI, including generative AI from outside providers. Whether your data trains anything, it doesn't say.

Beautiful.ai says its AI providers can't train on your decks and your data won't train public models. Whether its own models train on them, no document says.

Blink stores video and audio from cameras in your home. Its policy never uses "train", "AI" or "machine learning" — whether that footage builds anything, the document does not say.

Bumble sends your message prompts to OpenAI and tells you how long OpenAI keeps them. It does not say whether they train anything.

Calm holds what you use to sleep and what you tell it about your mood. Its privacy policy does not mention AI, machine learning or training anywhere.

Chase's online privacy policy never mentions AI, machine learning or model training. Its parent's institutional notice doesn't either.

Citi's US privacy notice for consumers never mentions AI or model training. Its notice for institutional clients promises not to train third-party models on their data.

Clue built its own foundation model for its AI features and says so. What that model was trained on, the policy does not say — though it is clear no data reaches Anthropic.

Coursera wrote a whole "How We Use AI" section listing nine AI features, and never says whether your work trains them. Its terms do forbid you from training on Coursera.

Cresta says it customises its models with customer data, with personal details redacted. Whether one company's calls train models that serve another, none of its documents says.

DBS mentions artificial intelligence twice, both times as an interface through which data is collected or used, never as what is built from it. On training, the policy says nothing.

Decagon's privacy policy never states whether your content is used to train AI models — a documented absence, not an unread document.

Depop's policy never uses the word "train". Its help centre says its recommendation systems "learn your interests" from what you view, like and buy. Neither says whether any of it trains a model.

Deutsche Bank's privacy notice contains no reference to artificial intelligence, machine learning or model training.

Cognition says it may use your content to train the models behind Devin — but makes it conditional on terms the policy doesn't reproduce.

D-ID says data awaiting processing isn't touched for model training. Asked what happens during processing, it answered — but in an email, not in the policy it pointed us to.

Discord never uses the word "train" and never says "AI". It does say it builds models from reported and widely available content, to detect prohibited conduct. Beyond that it does not say.

Etsy's privacy policy, in its UK and US versions, never mentions training, models or AI. What its automated processing and "improve our Services" reach is not stated.

Anker's policy for Eufy gives three training permissions, each gated on something you volunteer for. None of them covers the cameras, and about the cameras it says nothing.

Expedia's statement has a section on its use of AI: pricing, fraud, moderation and "feature generation" including embeddings. It never says whether any of it is trained on you.

ExpressVPN sells not keeping records. Its privacy policy makes no mention of AI, machine learning or model training either way.

Flo said in 2021 that it trains models on "Flo's large user population". Its FAQ now asks whether your data trains Flo's AI and answers only for third parties. Flo is being written to.

GitLab's AI terms mention training once: to forbid you from using GitLab Duo to train a competing model. Whether GitLab trains on your code isn't stated.

Gong records sales calls and optimises AI on them under legitimate interest. What the optimising is done with, the policy does not say.

The Guardian says it may use generative AI on personal data, and points to principles held on another page. What it may build from that data, this document does not say.

Headspace lists powering machine-learning algorithms among the uses of your personal information. Whether your data trains them, it never says.

Hotels.com names fourteen things it uses AI for, then adds "Other". Whether your data trains the models behind them is the one use it does not name.

HSBC's group privacy notice and its UK notice never mention AI, machine learning or model training. The one "train" in either is about staff.

Hugging Face says it does not train on user data, and publishes that only for its inference routing. For what you upload to the Hub, the privacy policy still says nothing.

Hypotenuse AI's terms reserve the right to use data about your content to improve its services and for other development. Neither they nor the privacy policy says whether that includes training.

J.P. Morgan's privacy notice does not mention artificial intelligence, machine learning or model training even once. A documented absence, at a firm that talks about AI constantly elsewhere.

Khan Academy contractually bars its AI providers from training on what you type. It does not say whether Khan Academy itself does.

Klarna's UK privacy page is an overview that defers, twice, to a Privacy Notice on a host that disallows this monitor. What the notice says about training has not been read here.

Krea's Business terms promise not to train on your inputs or outputs. The terms and privacy policy for Free, Basic, Pro and Max never say whether yours are trained on.

Manus's policy now says de-identified data made from your personal information may be used "to develop and improve our models", with an opt-out by email in some places. Manus is being written to.

Mullvad's two policies together run to 18,000 characters and never mention AI, machine learning or training. What they do say is that no profiling takes place at all.

Murf sells synthetic voice. Its privacy policy uses the word "voice" zero times, and "AI" only inside its own web address.

MyFitnessPal names machine learning and artificial intelligence once, as tools it uses to improve the Services. Whether your food and exercise logs train a model, it never says.

N26's policy has a section on its AI systems: fraud, credit risk, support, personalisation and "to improve our products and services". It never says whether any of them is trained on you.

Netflix's help centre says your viewing signals "continually update our algorithms". Neither it nor the privacy statement says train, model or machine learning. Netflix is being written to.

Nightfall's only training statement covers Google Workspace data. What it does with everything else isn't addressed.

Nintendo's privacy policy is 41 paragraphs and does not use the words "train", "AI" or "machine learning" once. Whether your play, voice chat or purchases train anything is not stated.

Nord's policy mentions machine learning once, for a bank-monitoring feature that reads transaction patterns, and never says whether anything of yours trains a model.

Oura's policy puts its own use of your data under a heading that says "Large Language Model Training", and rules out only selling it for somebody else's LLM. Whether it trains a model, it never says.

Outreach excludes Google Workspace data from AI training. Its own sales and email data goes unmentioned.

pCloud sells encrypted storage. Its privacy policy does not mention AI, machine learning or model training at all.

Quizlet's policy never mentions training, models or AI at all, although the product has an AI tutor. Whether your study sets and chats train anything is not stated.

Reddit's privacy policy names OpenAI's ChatGPT, and says third parties pay licensing fees for bulk content. It never once uses the word "train".

Regie.ai's policy said it did not use any data to train generalised AI models. The policy rewritten on 4 August 2026 dropped that section, and says nothing about training at all.

Replit's blog says public Repls are used for AI training, anonymised; the terms and privacy policy do not say so. This entry is being re-graded on that sentence, and Replit is being written to.

Ring's notice never says "train". Its US render links a disclosure: generative models trained on datasets that "in some cases" hold personal information. Whether your video is in them, neither says.

Samsung records your video chats with support "for training and quality control purposes" — the one phrase in 58,000 characters that could mean a model, and it is not explained.

Shopify explains its machine learning as automated decision-making with a human in the loop. What the models learn from, and whether that includes you, is a different question it does not answer.

Sonos's September 2026 statement sends 27voice commands to its cloud for "AI-assisted processing" and shares voice data with AI partners. It still never says whether any of it trains a model.

Stack Overflow guarantees its AI providers won't train on what you type, then takes a perpetual, irrevocable licence to it for "product development" without saying whether that includes training.

Starling's AI notice says no external AI model is trained on its customers' data. Whether its own models are, neither that notice nor its privacy notice says.

Valve's Steam privacy agreement never mentions training, models or AI. What happens to your reviews, chat and play data beyond recommendations is not stated.

Strong's privacy policy never mentions AI, machine learning or model training. Its only use of the word "training" means weights and reps.

Sync.com sells end-to-end encrypted storage. Its privacy policy mentions training once, and it is about staff.

Neither Systeme.io's privacy policy nor its terms say whether your content is used to train AI models. A documented absence, not an unread document.

Telegram's privacy policy never states whether your content is used to train AI models. This is a documented absence, not an unread document.

Temu's privacy policy never mentions AI or model training. The only "model" in it is your phone's.

Threema publishes two privacy policies and both of them govern its website. Neither describes what the messenger does with your messages. That is a documented absence, not a finding against the app.

Tinder names machine learning as a way it develops features, on the same wording as its sister app Hinge. Neither says whether your data trains anything.

Unity's player policy says it may use third-party AI services to run its business, and never says whether what it collects from you in games trains anything.

Vivaldi's browser privacy policy is short and direct, and it does not mention AI, machine learning or model training at all.

Waze's policy never mentions training, models or AI. Your routes go "to improve the Services or to develop new features" and to Google's maps, but whether any of that trains a model is not stated.

Wells Fargo's notice of data collection runs to 22,000 characters and mentions AI, machine learning and model training exactly zero times.

WHOOP's policy promises never to sell your data and sends you to another page for WHOOP Coach, the AI feature. Whether your data trains anything is answered there or nowhere.

Wise's assistant terms say your chat interactions are used to "refine the accuracy and improve the output" of an assistant built on third-party language models. Neither document says train.