Skip to main content

Privacy & Content Compliance

This page explains how the Turing Platform handles your data when you call the API, as well as content compliance requirements and violation handling procedures.

Data Privacy​

No Conversation Data Retained​

When calling model APIs through the Turing Platform, the platform performs only a single billing entry per request and does not memorize or retain any conversation content — including request inputs and model outputs.

The platform records only the metadata required for billing and usage statistics (call timestamp, model name, token count, request success status, etc.) for invoicing and usage analysis. Conversation content is never involved.

Upstream Provider Privacy Commitments​

The Turing Platform aggregates capabilities from multiple model providers. All integrated providers have signed privacy agreements with the platform, committing to:

  • Not using user data for model training;
  • Not disclosing user data or sharing it with unauthorized third parties.
Enterprise Data Compliance

If you have more specific data compliance requirements (data residency regions, enterprise data processing agreements, etc.), please raise them with the operations team during the enterprise onboarding process.

Content Compliance & Usage Policy​

Self-Governance for Integrators​

Integrators must self-govern and must not use the Turing Platform API to generate or request malicious, illegal, or harmful content, including but not limited to:

  • Hate speech, discrimination, and harassment;
  • Violence and content that endangers personal safety;
  • Pornography and content involving minors in violation of regulations;
  • Self-harm and suicide-related content;
  • Guidance or facilitation of illegal or criminal activity.

Content Moderation Mechanism​

All upstream model providers have built-in content moderation that identifies harmful content in real time throughout the call pipeline. Their classification approach aligns with mainstream industry standards (e.g., Azure OpenAI content filtering):

  • Categorized by harm type — hate, sexual, violence, self-harm, etc.;
  • Graded by severity — safe / low / medium / high;
  • Content that reaches the blocking threshold is directly rejected or filtered by the provider.

Since the platform retains no conversation content, moderation is completed in real time at the point of each call. The platform receives only the provider's moderation result flags and request-level usage characteristics; it does not read or store conversation content.

Violation Handling​

The platform identifies abuse based on moderation flags and call patterns — high-frequency high-risk content, repeated jailbreak attempts, and similar behaviors are classified as abuse and handled according to severity:

  • Minor / isolated violations: Alerts and rate limiting are triggered, and the integrator is notified to remediate;
  • Serious / repeated violations: The corresponding API Key is banned; in severe cases, the account is banned.
Risk of Account Suspension

Repeated or serious violations can result in your API Key and even your account being banned. Please implement robust content controls on your side — especially if your product serves end users — to prevent individual users' abusive behavior from affecting your entire account.