Legal

AI Policy

This AI Policy explains how AI is used in the Intelligence Platform (the "Service"), operated by Iridae ApS ("we", "us", "our"). Capitalized terms not defined here have the meaning in the Terms of Service.

Scope

This policy covers:

  • AI systems used to generate Outputs
  • human review in the AI workflow
  • AI-related subprocessors (model providers and inference hosting)
  • training/improvement of our systems using Customer Content

AI systems used

We generate Outputs using a combination of:

  • first-party models (developed and operated by us)
  • third-party models (external model providers)

We may route a request between these systems for quality, safety, latency, or availability.

Human review

Some requests may be processed and/or reviewed by humans to prepare inputs, quality-check Outputs, or handle support and safety issues. Human access is restricted and subject to confidentiality obligations.

AI subprocessors

Customer Content may be processed by third parties strictly to provide AI functionality and underlying infrastructure.

Inference hosting for our first-party models (cloud/compute):

  • Amazon Web Services
  • Microsoft Azure
  • NVIDIA

Third-party model providers (where used):

  • OpenAI
  • Anthropic

We keep the current subprocessor list (including purpose and, where relevant, regions) at: /legal/subprocessors/

Cloud-hosted inference for our models

When our first-party models run on third-party infrastructure for inference, Customer Content is processed on that infrastructure to produce Outputs. We configure and operate the Service to minimize exposure (for example, limited access and operational safeguards). Some limited operational logging may occur for reliability and security.

Training and improvement

Unless you have a written no-training agreement with us, you agree we may use Customer Content to improve the Service, including training our systems.

We aim to do this in a privacy-preserving way, including approaches where learning is captured as model updates/weights rather than sharing your raw Customer Content across customer accounts. Even so, no approach guarantees zero risk of unintended memorization or leakage.

If your organization requires a no-training arrangement, you must obtain a written agreement from us before using the Service.

Updates and contact

We may update this AI Policy from time to time.

Questions: support@iridae.com