1. Is OpenAI using your data to train models?
- APIs
- No: prompt other data you provide is not used. Your data is not stored or used for training. API interactions are not retained beyond processing the request.
- Interactions may be reviewed in aggregated/anonymized form to improve models unless specified otherwise (e.g., enterprise settings).
- ChatGPT
- It’s a matter of your subscription:
- the Enterprise subscriptions (Team & Enterprise) are by default opted out
- the end-users subscriptions (including Plus) are by default opted-in
- HOWEVER, there is a setting in Settings in ChatGPT Plus that allows you also the opt-out (I have Plus and I can turn it off)
- Read more in OpenAI’s statement: How your data is used to improve model performance
- Temporary Chats are not used for training purposes.
- It’s a matter of your subscription:
- Reference:
- OpenAI’s Data Policy : As of March 1, 2023, data sent to the OpenAI API is not used to train or improve OpenAI models unless you explicitly opt-in. To help identify abuse, API data may be retained for up to 30 days before deletion (unless required by law). For trusted customers with sensitive applications, zero data retention may be available. With zero retention, request and response bodies are not logged and exist only in memory for processing.
2. Is Anthropic using your data to train models?
Anthropic defines:
- Consumers: Claude.ai free or Pro users
- Commercial customers: API, Console, Enterprise & Teams users
API Data Handling
- No Data Retention for Training:
- Inputs and outputs from API calls are not used to train future models.
- Anthropic does not store API request data beyond what is necessary for immediate processing.
- Enterprise users can be assured that their data remains private and is not accessed beyond execution.
- Security & Encryption:
- There is no manual review of API requests unless flagged for security reasons.
- API data is not accessed by employees, except in rare cases where intervention is needed for security purposes.
- Logs are minimized and automatically deleted after processing.
- No Data Retention for Training:
Chat (Consumer Chatbots like Claude AI)
- Default Data Retention Policy:
- These chats are not used to train AI models unless a user explicitly opts in.
- By default, Anthropic may store chat interactions for up to 30 days for security and abuse monitoring.
- BUT your data is used when given Explicit Feedback (e.g. Thumbs Up): When you provide explicit feedback, such as using the thumbs up/down feature or reporting materials directly, Anthropic may use that specific interaction to improve their models. (see privacy.anthropic.com)
- Opting In: Your data will only be used for training if you have explicitly opted in to such use. (see privacy.anthropic.com)
- Content Review for Safety & Compliance:
- Some chat interactions may be reviewed by Anthropic’s Trust & Safety team if flagged for violations (e.g., abuse, misuse). However, general conversations are not monitored in real time.
- Default Data Retention Policy: