Working name · aipolicylog.com
Does this AI tool train on your data?
Dated quotes from official vendor pages on whether AI tools train on your data. Each row is a vendor plan, quoted from the official page we checked on 2026-11-18 in America/Toronto (UTC−5). We do not issue verdicts.
Not legal advice. Training, retention, and zero data retention are different things. Consumer and business plans often differ. If a field was not on the pages we read, it is marked “Not stated / not verified.” Policy pages change; anything material waits for human approval before it is published.
| Tool | Plans covered | Trains on inputs by default? | Page status | Checked |
|---|---|---|---|---|
| OpenAI / ChatGPT | 4 plans | Consumer Free/Plus/Pro: sharing is on by default, with an opt-out. Business, Enterprise, Edu, Healthcare, Teachers, and the API: not used for training by default; API opt-in exists. | complete | 2026-11-18 |
| Anthropic / Claude | 2 plans | Consumer (Free, Pro, Max, Ultra): may be used when the model-improvement setting is on, with an opt-out. Commercial (Team, Enterprise, API, Claude Gov): not used by default unless the customer opts in (for example the Development Partner Program). | complete | 2026-11-18 |
| Google Gemini | 2 plans | Gemini Apps (consumer): training depends on Gemini History settings; temporary chats are not used to train. Google Workspace: customer data is not used to train models without the customer’s prior permission or instruction. | complete | 2026-11-18 |
| Microsoft Copilot | 3 plans | Microsoft 365 Copilot and the current (post–August 18, 2026) consumer Copilot app: official pages say prompts, responses, and file contents are not used to train foundation models. An older consumer FAQ, which Microsoft says applies only to the pre-update app, described an opt-out for model training. | complete | 2026-11-18 |
| GitHub Copilot | 2 plans | Free, Pro, Pro+, and Max: may be used to train models unless you opt out (effective April 24, 2026). Business and Enterprise: not used to train models, under GitHub’s Data Protection Agreement, without customer authorization. | complete | 2026-11-18 |
| Cursor | 2 plans | If Privacy Mode is on, Cursor says it will not train on your data and maintains ZDR agreements with providers. If Privacy Mode is off, Cursor says it may store data and train models. Privacy Mode is available to anyone (free or Pro). | complete | 2026-11-18 |
| Perplexity | 1 plan | Consumer Privacy Notice (Oct 23, 2026): Perplexity says it uses data to improve or create services and products, including its AI models, and that it processes prompts and queries. Email content is not used to train models. Enterprise/API rules were not verified (enterprise page 404; notice carves those offerings out). | partial | 2026-11-18 |
| Notion AI | 2 plans | By default, neither Notion nor its AI providers use Customer Data to train models. Most models offer ZDR; non-ZDR models are off by default and require a workspace-owner opt-in. A BAA is described as available to Enterprise customers. | complete | 2026-11-18 |
| Slack | 1 plan | Slack says it will not use Customer Data to train generative AI models unless the customer provides affirmative opt-in consent. A separate opt-out exists for Slack “global models” (non-generative ML) via a request to feedback@slack.com. | complete | 2026-11-18 |
| Zoom AI | 1 plan | Zoom says it does not use customer audio, video, chat, screen sharing, attachments, or other communications-like customer content to train Zoom’s or its third-party AI models. Third-party providers are described as under zero-data-retention policies, with limited trust-and-safety exceptions. | complete | 2026-11-18 |
| Apple Intelligence | 1 plan | Private personal data and user interactions are not used to train foundation models unless you explicitly opt in (for example Improve Siri & Apple Intelligence). Private Cloud Compute is described as not retaining request data. | complete | 2026-11-18 |
| Amazon Bedrock / Amazon Q | 1 plan | Model providers do not have access to Bedrock customer prompts and completions, per the data-protection guide. Retention is configurable, including a zero-data-retention mode. A blanket AWS-training statement from the legal disclosure page was not fully verified in this pass. | partial | 2026-11-18 |
| Meta AI | 1 plan | Official Meta pages say training/improvement data for AI at Meta includes people’s interactions with AI at Meta features (content and messages exchanged with those features), plus public and licensed data. A dedicated interaction-training opt-out was not verified on the pages we read. | partial | 2026-11-18 |
| Grok (xAI) | 2 plans | Logged-in users can choose whether User Content is used to train models. When not logged in, the terms say you grant full rights to use data for training, where permitted. Default toggle state for logged-in users was not explicitly labeled on the legal pages we read. | partial | 2026-11-18 |
| DeepSeek | 1 plan | The official English privacy policy (updated February 10, 2026) lists training and improving machine-learning models as a use of Personal Data, including user inputs, and mentions a right to opt out of using Personal Data for training. Default control state and retention days were not clearly numbered. | partial | 2026-11-18 |
| Mistral AI | 1 plan | The official consumer privacy policy (effective September 3, 2026) lists training models on Inputs and Outputs as a purpose and provides an in-account control to object to that use. Input/Output are kept 30 rolling days to monitor abuse unless ZDR is activated, except for specific APIs. | partial | 2026-11-18 |
| Grammarly | 1 plan | Grammarly offers an opt-out of using your content to train its models (Product Improvement and Training). Third-party providers are not allowed to train on user content. Default toggle state was not explicitly labeled on the pages we read. | partial | 2026-11-18 |
| Otter | 1 plan | Otter’s privacy policy says it trains proprietary AI on de-identified audio recordings and on transcriptions, which may contain Personal Information. A model-training opt-out was not clearly stated on the pages we read. | partial | 2026-11-18 |
| Adobe Firefly | 1 plan | Adobe’s Firefly approach page says customer data and content are never used to train generative AI models. The General Terms of Use say Adobe does not train generative AI models on your or your customers’ content unless the content was submitted to the Adobe Stock marketplace. | complete | 2026-11-18 |
| Canva | 2 plans | Teams, Business, Enterprise, and Education user content is not used to improve AI-powered features (Education: “We will not use User Content of Canva Education for AI training”). Individual accounts are directed to privacy controls to manage use of data for AI training. | complete | 2026-11-18 |
| Salesforce Agentforce | 1 plan | The official Agentforce Privacy FAQ (published June 2025) says Salesforce does not currently use Customer Data to train generative AI models, and that third-party LLMs are under a zero-data-retention policy so they do not retain Customer Data or use it for training. Future training would require a change to Product Terms with prior notice, per that FAQ. | complete | 2026-11-18 |
| Intercom / Fin | 1 plan | Third-party AI providers are contractually restricted from training on Customer Data and are described as having zero retention once a response is generated. Fin may use anonymized customer data for Intercom’s own fine-tuning, with an opt-out and deletion within 30 days. | complete | 2026-11-18 |
| Together AI | 1 plan | Training is a separate opt-in and is off by default. ZDR can be enabled at organization level so prompts and outputs are not stored. Unless ZDR is enabled, Together stores prompts and responses and may use them for product improvements (which the docs distinguish from training). | complete | 2026-11-18 |
| OpenRouter | 1 plan | OpenRouter is an aggregator. It does not claim a single training rule for all models: “We do not control, and are not responsible for, LLMs’ handling of your Inputs or Outputs, including for use in their model training.” OpenRouter says it does not use uploaded-file content to train its own models. Provider logging and training depend on the selected model. | complete | 2026-11-18 |
| Figma AI | 3 plans | Sharing customer content for Figma’s own model training is optional and admin-controlled: on by default for Starter and Professional, off by default for Organization and Enterprise, and not used for Figma for Education (K-12) or Figma for Government. Figma says AI service providers are prohibited from training on Service data. | complete | 2026-11-18 |