AI Disclosure & Fictional Content

When you question a suspect, an AI model produces the suspect’s reply in real time. The model is given a persona sheet, the allowed and forbidden things this suspect knows, and, when you present evidence, the description of the contradiction it must acknowledge if you have proven one.

The AI writes the words. It does not choose the mystery.

• The AI does not decide who the culprit is, what the motive was, or how the crime was committed. Those facts are canonical, stored server-side, and not exposed to any AI prompt.
• The AI cannot promote you, alter your score, unlock evidence, add contradictions, or change the verdict. Only the deterministic case engine can.
• The AI never sees your name, email address, or purchase history.
• The AI’s output passes through a server-side guard that strips unsafe content and blocks accidental spoilers before it reaches you.

Suspect dialogue is currently produced by an Anthropic Claude model, called through a managed integration. Prompts and replies are not stored in our analytics; only aggregate latency and error rates are recorded so we can keep the service reliable.

Every case in CaseZero is a work of fiction. All persons, companies, publications, locations and events depicted are invented. Any resemblance to real people, real organisations, or real crimes is unintended and coincidental.

Rarely, the AI may produce output that is confusing, off-tone, or appears to contradict the case. When this happens, the deterministic state is still correct: what you have unlocked, what contradictions you have proven and the final verdict all come from the server, not from anything the AI says.