AI Addendum

1. No cross-tenant training

Cavaridge does not train models on your data. Every AI request is scoped to your tenant. We do not aggregate, anonymize, or otherwise reuse your prompts or completions for model training.

2. Sub-processor flow-down

Cavaridge AI calls underlying model providers (Anthropic, OpenAI, Google, Perplexity, others available through OpenRouter). Each provider's training-data policy applies to traffic routed to that provider; Cavaridge selects providers with no-training defaults and verifies via the OpenRouter routing config.

3. Healthcare

For Healthcare-tier tenants, Cavaridge routes only to model providers with active BAAs. The current eligible-provider list lives at /legal/security and is verified per-request by the Cavaridge AI gateway.

4. Outputs

AI-generated content is yours. Cavaridge AI may surface a "this output was AI-generated" badge in artifacts; you control whether that badge appears in shared / published artifacts.

5. Hallucination + accuracy

AI outputs may be inaccurate, biased, or outdated. Cavaridge AI surfaces source citations where applicable (search-grounded routes). For high-stakes use (legal, medical, financial), users must verify outputs before acting.

6. Spend controls

Cavaridge enforces per-tenant LLM spend caps server-side. Caps are set in your tenant config; runaway costs are not possible.