Model training
Using prompts, files, or responses to develop or improve a provider’s models. Provider policies usually make this an account-level or contractual choice.
Privacy record · AI providers
Data practices
stay visible.
How the AI services used by EX MULTIS describe model training, temporary retention, and Zero Data Retention controls—linked to the provider materials that govern each summary.
LAST REVIEWED · SEPTEMBER 4, 2026OFFICIAL SOURCES · 6 PROVIDERS
The operating position
Training and retention are different controls.
EX MULTIS does not direct AI providers to use customer content to train their models. Most commercial API providers say training is off by default. That does not always mean the provider keeps no temporary records.
Retention depends on the provider, account, endpoint, and features used. This page keeps those distinctions visible rather than treating a single label as a guarantee.
Read the record correctly
Data retention
Keeping request or response content temporarily for security, abuse monitoring, troubleshooting, or a feature that depends on stored state.
Route controls
The provider account, exact endpoint, contractual terms, and available technical confirmation must agree before a stricter retention mode is represented as active.
Provider record
Current published positions
The linked provider documents control if a summary and its source differ.
01Not used by default
OpenAI
Model review, verification, and image generation
Model training
API inputs and outputs are not used to train or improve models unless the customer explicitly opts in.
Retention
Standard API use may include abuse-monitoring logs retained for up to 30 days. Modified Abuse Monitoring or Zero Data Retention may be available for approved organizations; endpoint exceptions apply.
Official sourcesAPI data controls ↗Enterprise privacy ↗
02Not used by default
Anthropic
Independent model review
Model training
Commercial API content is not used for model training unless the customer participates in a development program or submits qualifying feedback.
Retention
Inputs and outputs are generally deleted within 30 days. Longer periods can apply to feedback, policy enforcement, legal requirements, or features that require storage.
Official sourcesModel-training policy ↗Commercial retention ↗
03Prohibited by API terms
Perplexity
Current-source research through Sonar and Agent API routes
Model training
Its API terms state that customer content is not used, or authorized for use, to train, retrain, fine-tune, or otherwise improve generative AI models.
Retention
Perplexity publishes a strict Zero Data Retention policy for Sonar. Its public Sonar privacy page does not make the same endpoint-specific statement for newer Agent API routes.
04Only with permission
xAI
Grok research and comparison
Model training
xAI states that it never trains on API inputs or outputs without the customer’s explicit permission.
Retention
Requests and responses are stored for 30 days by default for abuse auditing. Team-wide Zero Data Retention can remove that content retention but disables features that require provider-side storage.
Official sourcesAPI security and ZDR ↗
05Depends on billing tier
Google Gemini
Model review and document extraction
Model training
For paid Gemini API services, prompts and responses are not used to improve products. For unpaid services, Google may use submitted content and responses to improve products and machine-learning technologies.
Retention
Paid-service prompts and responses may be logged for a limited period for safety, security, and required disclosures. Grounding with Google Search has an additional 30-day storage term.
Official sourcesGemini API terms ↗Billing and plan check ↗
06Not used by default
Mistral
Document OCR
Model training
API data is not used for model training by default. Administrators can control data use, while Labs models may follow different training terms.
Retention
Zero Data Retention is available on eligible paid plans for supported stateless endpoints, including the OCR endpoint used for document extraction.
Official sourcesPrivacy and data controls ↗Zero Data Retention ↗
Verification discipline
Public policy is the starting point, never the whole control.
Provider account settings, the exact endpoint used, contractual terms, and any available technical confirmation must also agree before EX MULTIS represents a stricter retention mode as active.
- 01Training or data-sharing options remain disabled unless a customer gives separate, informed authorization.
- 02Zero Data Retention is described as active only after the provider account or contract confirms it.
- 03If a requested data-handling level is incompatible with a provider or feature, the route is restricted rather than silently downgraded.
Before sharing sensitive information
Use approved safeguards for approved work.
Do not submit secrets, regulated records, or highly sensitive personal information unless your organization has approved the service and an appropriate written agreement is in place.
Provider policies and product features can change. EX MULTIS reviews this page periodically and records the review date above. See the full Privacy Policy for EX MULTIS’s own handling of account information, content, and operational records.