AI Use

We use first-party Claude (Anthropic) models for in-product features. We do not use third-party model providers without legal review.

Customer data is never used to train AI models. Provider contracts include zero-retention / no-training clauses for all customer prompts and outputs.

When a customer-facing feature sends content to an AI provider, we record the event and the destination in our access logs. Customers can opt out of AI-assisted features at any time via account settings.

For high-stakes AI-generated content (compliance summaries, security recommendations) we require human review before publication. Lower-risk features (search, summaries) are generated and shown without manual review, with a visible "AI-generated" disclosure.

An internal AI committee reviews our model providers, data flows, and customer-facing AI features quarterly. Decisions are logged in our governance system.