AI-Generated Content Disclosure

Every case in CaseScene — the victim, the suspects, the timeline, the weapon, the motive, the closing reveal — is generated by Anthropic's Claude model family (currently Claude Sonnet for case authoring and Claude Haiku for live interrogation). No two players receive an identical mystery. A small human-edited library of cases is provided as a baseline, but the bulk of content you encounter is composed at request time.

When you interrogate a suspect, you are conversing with a Claude-powered character driven by a private dossier the model never reveals verbatim. Replies are produced fresh each turn. Suspects may bluff, contradict themselves, or lie within the rules of the scenario — this is intentional. They will not produce explicit, sexual, hateful, or illegal content.

Your prompts, your interrogation transcripts, your account information — none of it is used to train AI models. We call the Anthropic API with zero-retention configuration where available. Game telemetry stored in Firebase is used only to operate the service, not to retrain any model.

All AI calls originate from our server. The ANTHROPIC_API_KEY is never exposed to the client. Requests are rate-limited per session and per account. A three-tier fallback (Cloud Functions → direct API → curated demo case) keeps the game playable when a provider has an outage.

Generated cases pass a content filter before they reach you, and a human review queue receives any case flagged for safety, factuality, or quality. You may report a case directly from the result screen and it will be triaged within 72 hours.

You can request deletion of your account, your transcripts, and any case generated for you. Write to privacy@casescene.app and we will respond within 30 days. For Turkish residents this is also the contact point for KVKK Article 11 requests.