Does Your AI Vendor Train on Your Data? 10 AI Platforms Compared
Each row is the training clause published for that vendor. A blank figure means the vendor has not published it.
Last refreshed 21 Oct 2026.
Next refresh Mon 26 Oct at 4:00 AM PT.
| Vendor | Trains on your data | Published clause |
|---|---|---|
| AWS Bedrock | No training by default | Inputs and outputs are not used to train Nova, Titan, or third-party models. |
| Azure AI Foundry | No training by default | Prompts and completions are not used to train foundation models without your permission. |
| Google Cloud Vertex AI | No training by default | Customer data is not used to train or fine-tune models without prior permission. |
| Databricks (Mosaic AI) | Varies by tier | Paid Model Serving does not train on inputs or outputs. Other terms may use inputs that do not identify the customer. |
| Snowflake (Cortex AI) | No training by default | Usage and customer data are not used to train models made available to other customers. |
| OpenAI (API & Enterprise) | No training by default | As of 1 March 2023, API data is not used to train or improve models unless you explicitly opt in. |
| Anthropic (Claude API & Enterprise) | No training by default | Commercial terms say Anthropic may not train models on Customer Content from Services. |
| Cohere (Enterprise Platform) | Varies by tier | Private deployments are not accessed. SaaS training depends on the Data Controls opt-out. |
| Mistral AI (La Plateforme) | Varies by tier | Training is off except for the opt-in, opt-out, Feedback, Order Form, and Labs or Preview cases in section 4.2. |
| xAI (Enterprise API) | No training by default | Business inputs and outputs are not used to train models unless you agree, including the free-credits case on the FAQ. |