Responsible AI at OQANDA

OQANDA uses AI to help moderators do their job better. AI on OQANDA is always an assistant — the human decides. This page explains what that means in practice.

Four principles we do not break

  1. Language of origin is sacred. Every message keeps its original language. Translation, when we offer it (v2.0), is always labeled and the original is always one tap away.
  2. Translation is always marked as translation. Never silently substituted.
  3. AI never acts without human authorization. Scores suggest. Moderators decide.
  4. Nothing is ever deleted or out of reach. Archives stay archives.

What AI can do on OQANDA

  • Suggest a quality or relevance score for each incoming question
  • Group duplicate or near-duplicate questions so moderators do not read the same question ten times
  • Flag likely spam for moderator review
  • Pre-sort questions in a defined emergency, when no human moderator is available — with the guardrails below

What AI cannot do on OQANDA

  • Delete your question
  • Hide your question from the moderator queue
  • Publish your question without human approval
  • Activate Emergency Moderation Mode on its own
  • Keep running in Emergency Mode indefinitely

Emergency Moderation Mode — the details

If every human moderator becomes unavailable during a live Event, an Administrator can temporarily authorize AI-assisted moderation.

  • Authorization is manual and per-event. No auto-activation, ever.
  • Time-capped: 30 minutes per activation, with a hard maximum of 105 minutes. After that, moderation stops and waits for a human.
  • Held-for-human bucket. Borderline cases — roughly the top 10% of ambiguous questions — are held for human review rather than auto-routed.
  • Transparent banner. Every participant — fans, VIPs, the artist — sees a banner when Emergency Mode is active. Nothing about it is hidden.
  • Moderator return is explicit. When a human moderator returns, they click "I'm here" and Emergency Mode stops.

Administrator kill switches

An Administrator can disable AI entirely — or disable any individual component (quality scoring, duplicate clustering, spam flagging, sentiment, or emergency mode) — at any time, during live operation. Disabling is instant. No scores are computed. No calls leave our servers for AI purposes. Moderation returns to fully human-driven.

Where your data is processed

AI moderation runs on Google Cloud Vertex AI, hosted in europe-west4 — a European Union region. Your content is processed transiently to return a score or a cluster, and is not retained by Google.

Your data is never used to train Google's AI models. This is enforced by configuration, not just by policy.

What changes in v2.0

Multilingual participation — reading and writing in your own language, with translation always labeled and the original always one tap away — is planned for v2.0, not the initial launch. At launch, AI helps moderators. In v2.0, AI also helps fans speak across languages. The four principles above apply to both.

Questions or concerns

Contact privacy@oqanda.com. For the full legal text, see our Privacy Policy.