← All pages

Competing Model Ban

Definition: A clause banning a user, developer, or customer from using a provider’s Input, Output, or Services to build, train, or improve a competing product or AI model — moat-protection language layered on top of the ordinary Input/Output license grant. Risk level: Moderate-to-high. Unlike most clauses in this wiki, this one restricts the user’s downstream rights instead of expanding the provider’s rights over user data — but it’s predatory in a different way: it can trap a developer’s own work (built using a provider’s Output) inside that provider’s ecosystem, since “improving a competing model” is often worded broadly enough to cover ordinary fine-tuning or distillation.

Note on provenance: This concept page covers the sources/GenGA/ (GenGA/Open Terms Archive) dataset, which has no pre-tagged JSONL — all tagging here is LLM-assigned by direct reading. See each linked platform page’s methodology note.

Platforms Using This Clause

Common Wording

Using Output to develop models that compete with OpenAI. — ChatGPT, Terms of Service, 2025-11-11

[Commercial Terms §3.3(e)-equivalent ban on using Output/App Requests “to develop models that compete with OpenAI.”] — ChatGPT, Commercial Terms / Developer Terms, 2025-11-11

Use Restrictions. Customer may not and must not attempt to (a) access the Services to build a competing product or service, including to train competing AI models or resell the Services except as expressly approved by Anthropic; (b) reverse engineer or duplicate the Services; or (c) support any third party’s attempt at any of the conduct restricted in this sentence. — Claude.ai, Commercial Terms, 2025-11-11

To develop any products or services that compete with our Services, including to develop or train any artificial intelligence or machine learning algorithms or models or resell the Services. — Claude.ai, Terms of Service, 2025-11-11

Output Restrictions. To the extent permitted by applicable law, Customer may not use image Outputs to develop or train any image generation product that competes with a Mistral AI Product. — Le Chat, Commercial Terms, 2026-03-12

Limits on use of data from the AI Services. You may not use the AI services, or data from the AI services, to create, train or improve (directly or indirectly) any AI technology. — Microsoft Copilot, Acceptable Use Policy, 2026-02-18

Create or make available any products or services competitive with, similar to, or that would otherwise be a substitute for the Services (or the third-party AI models used in the Services), except with Perplexity’s express prior written permission. — Perplexity, Acceptable Use Policy, 2025-11-11

Suspension of Customer Applications. …if Perplexity determines that a Customer Application is… competitive with Perplexity… Perplexity may terminate the license… and deactivate the relevant API Key(s). — Perplexity, Developer Terms, 2025-11-11

use any information, data, or content (including without limitation the Outputs) made available via the Services to develop or improve any products or services (including without limitation any models) that compete with or are similar in functionality to the Services (or any aspect thereof). — Qwen Chat, Terms of Service, 2025-11-11

(b) use any Service to help develop, or help provide to any third party, any product or service similar to or competitive with any Service. — xAI, Commercial Terms, 2025-11-11

Using the Service or any Output to develop models or services that compete with xAI, scraping or reselling any Input or Output, or distilling model data. — xAI, Terms of Service, 2025-11-11

Notes & Trends

OpenAI and Anthropic both ban all competing-model training on their Output/Services, using near-identical phrasing (“to develop models that compete,” “to train competing AI models”) across each provider’s Commercial Terms, Developer Terms (OpenAI only), Terms of Service, and (Anthropic only) Acceptable Use Policy — a deliberate, company-wide moat-protection stance. Mistral’s Le Chat is the outlier: its ban is scoped narrowly to image Outputs training image-generation competitors, and even that narrow restriction shrank further — an earlier (2025-12-15) version of Le Chat’s Commercial Terms also banned using web-search Outputs “as part of any machine learning or similar algorithmic activity” or to “create, train, evaluate, or improve commercial products,” but this broader restriction was removed by 2026-03-12, leaving only the narrow image-generation ban. Microsoft Copilot’s ban is the broadest of all: rather than limiting itself to competing products, it bans using the AI services or their data “to create, train or improve… any AI technology” — wide enough to cover unrelated, non-competing AI development. Perplexity’s ban is broader than OpenAI’s/Anthropic’s in a different way — it covers competitive products and services generally, not just training activity, and is enforced two ways: a flat usage prohibition (Acceptable Use Policy) and a suspension/termination right (Developer Terms). Qwen Chat’s and xAI’s bans closely track OpenAI’s/Anthropic’s wording (“similar in functionality to the Services,” “similar to or competitive with any Service”) and are each paired with one of the dataset’s broadest Content licenses — an unusual combination of expansive provider rights and restrictive user rights in the same document, now confirmed for 2 of 11 GenAI providers. DeepSeek is the explicit inverse of this pattern (see its Terms of Service / Developer Terms pages): it affirmatively permits using Input/Output to “train other models (such as model distillation)” — the only provider in this dataset with no competing-model ban at all.