← All pages

Artificial Intelligence

Definition: A clause that explicitly mentions “artificial intelligence” when describing how a platform — or third parties it shares data with — collects, processes, or uses user data, most often to train AI/ML models or to moderate AI-generated media. Risk level: These clauses are the clearest, most citable evidence that a platform uses data for AI training — but how much control they leave users varies widely. Some are a flat first-party admission with no opt-out; some are a content-moderation rule that actually protects users; some disclose that third parties outside the platform’s control may train AI on shared data.

Platforms Using This Clause

Common Wording

media depicting a real person that has been fabricated or simulated, especially through use of algorithms or broader artificial intelligence. — X, Community Guidelines, 2025-07-22 (part of the “Misleading Media” manipulated-content rule)

X may use information that individuals provide and data that it receives (as described in X’s Privacy Policy) to train machine learning and artificial intelligence models, including generative models. This includes public X posts and associated metadata of X users. — X, Other (Legitimate Interests Analysis annex), 2024-07-25

We may use the information we collect and publicly available information to help train our machine learning or artificial intelligence models for the purposes outlined in this policy. — X, Privacy Policy, 2024-10-17 (present unchanged in every scrape through 2026-01-16)

[T]he recipients of the information may use it for their own independent purposes… including, for example, to train their artificial intelligence models, whether generative or otherwise. — X, Privacy Policy, 2024-10-17 (third-party-AI-training disclosure; present unchanged in every scrape through 2026-01-16)

You agree that this license includes the right for us to (i) analyze text and other information you provide and to otherwise provide, promote, and improve the Services, including, for example, for use with and training of our machine learning and artificial intelligence models, whether generative or another type… — X, Terms of Service, 2024-10-17 (the contractual license-grant basis for the same training right the Privacy Policy discloses; trigger verbs broadened to include “input, create, generate” starting 2025-12-17)

We support research in areas like artificial intelligence and machine learning to do things like create COVID-19 forecasting models… — Facebook, Privacy Policy, 2022-12-16 (baseline; present in substance throughout) — an AI-research-funding disclosure, not a first-party AI-training-on-user-data admission like X’s.

Generative AI models that make our features work may not have the most complete or recent information… Meta sometimes shares the messages you send to our AIs and general information, like your region, with these companies so you can get better results. — Facebook, Privacy Policy, 2024-11-15 (new “AI integrations” section) — discloses that messages sent to Meta’s generative AI features may be shared with third-party AI partner companies, a data-sharing disclosure distinct from X’s training disclosures.

We use and develop advanced technologies - such as artificial intelligence, machine learning systems, and augmented reality - so that people can use our Products safely regardless of physical ability or geographic location. For example, technology like this helps people who have visual impairments understand what or who is in photos or videos shared on Facebook or Instagram. — Facebook, Terms of Service, 2022-07-11 (present in substance throughout) — frames AI/ML around accessibility and safety, the narrowest framing of any platform’s Terms of Service tracked so far; no AI-training-on-Content language appears in this document (compare with the much broader disclosures in Facebook_Privacy_Policy.md).

Technologies like artificial intelligence and machine learning give us the power to apply complex processes across our Service. Automated technologies also help us ensure the functionality and integrity of our Service. — Instagram, Terms of Service, 2022-07-11 (present in substance throughout 128 scrapes) — a third, distinct framing: broad and generic (“apply complex processes,” “ensure functionality and integrity”), unlike Facebook’s accessibility-specific framing or X’s training-on-Content framing — no specific disclosure of what data trains what model.

Pinterest also has features that are supported by generative artificial intelligence technology. — Pinterest, Privacy Policy, 2025-05-01 (confirmed genuine via direct diff against the 2024-03-13 baseline) — a brief generative-AI-features disclosure directly adjacent to the platform’s training-on-user-content admission (see Train AI/Models); most of this keyword’s other captures on this page are incidental surrounding privacy-rights text (opt-out instructions, legitimate-interest disclosures) rather than AI-specific language.

…we also develop and improve the algorithms and machine learning models (an expression of an algorithm that combs through significant amounts of data to find patterns or make predictions) that make our features and Services work, including through generative AI features (artificial intelligence capable of generating text, images, or other media, using generative models…). For example, our algorithms and machine learning models take into account the conversations Snapchatters are having with My AI to improve the responses from My AI. — Snapchat, Privacy Policy, 2024-01-22 (present in substance throughout 8 scrapes) — names a specific in-house product, “My AI” (Snapchat’s chatbot), and discloses that user conversations with it are used as model-improvement signal — the most product-concrete first-party training admission under this keyword besides X’s and LinkedIn’s.

Notes & Trends

This concept was, for a long time, documented only on X, across four document types — and the wording differs across all four. X’s Privacy Policy carries the most direct first-party admission: it trains X’s own models on public posts and user data, with no AI-specific opt-out shown in the captured snippets. X’s Terms of Service supplies the contractual mechanism for that same training — the user-Content license explicitly lists AI/ML training as a permitted use, and as of 2025-12-17 that grant also covers content a user inputs into or generates with X’s tools, not just content they post. X’s “Other” bundle adds a third angle: a GDPR-style “legitimate interests” justification for the same training, plus (separately, see train AI/models) a clause banning third parties from doing the same via the API. X’s Community Guidelines use is narrower and more protective — a content-moderation rule about AI-manipulated media, not a data-use disclosure. Facebook’s Privacy Policy adds a fifth angle: a sprawling, multi-subsection AI-disclosure architecture (482 unique artificial intelligence-tagged paragraphs) covering AI-research funding, content-moderation classifier training, object-recognition training on user media, and third-party AI-partner data sharing — broader in volume than X’s, though spread across the document rather than concentrated in one or two clear clauses. Facebook’s own Terms of Service, by contrast, is the most conservative AI framing in the wiki — purely accessibility/safety-oriented — with the real breadth of Meta’s AI-training disclosures living in its Privacy Policy instead. Instagram’s Terms of Service breaks from Facebook’s pattern entirely: it’s an independently drafted document (confirmed zero byte-exact snippet overlap with Facebook’s ToS) using a generic “we use AI/ML to improve the Service” framing, with neither the accessibility specificity of Facebook’s ToS nor the volume of Facebook’s Privacy Policy. Pinterest’s Privacy Policy is the first case in this wiki where most of the captured artificial intelligence-tagged text is incidental — adjacent privacy-rights/opt-out sections swept into the same capture window as the one genuine generative-AI sentence — a reminder to check each capture’s actual content rather than assume the keyword tag always lands on AI-specific language. Snapchat’s Privacy Policy returns to a clean, on-topic, product-named disclosure: unlike Pinterest’s stripped-down reference, Snapchat explicitly names “My AI” and discloses that conversations with it directly feed back into improving its own responses — among the most concrete chatbot-training admissions in the wiki.