Murder mysteries for ChatGPT, Claude & Gemini
Updated
Whodunits whose solution is cryptographically sealed before you start — the AI can't cheat. Free, no signup. Each game is a ready-to-play prompt (we call it a game card) that you open in your own AI assistant.
A sealed mystery is a whodunit whose solution is committed as a SHA-256 hash before the game ships. The culprit, the method and the evidence are fixed in a file on this site, so the assistant is refereeing a case that already has an answer instead of deciding who did it once you accuse someone.
New to AI games? Read AI murder mystery games that can't cheat.
| Game | What it is | Players | Length | Ages |
|---|---|---|---|---|
| Blackfile №1: The Vanishing at Larchmont Hall | A 1925 locked-room whodunit: an heiress bolts a glass conservatory from the inside and is not there when the doors come down. | 1 | 60–90m | 12+ |
| Blackfile №2: A Death at the Wyke Hydro | A snowbound 1929 poisoning whodunit: six suspects, one cup of hot milk, and an answer sealed before you start. | 1 | 75–105m | 12+ |
| Blackfile №3: The Long Exposure at Harrowfell | A 1953 observatory whodunit: a death on the dome stair, six suspects who cannot leave the fell, twenty moves, and a culprit sealed before you start. | 1 | 60–90m | 12+ |
Can ChatGPT run a murder mystery without cheating?
The failure mode of playing detective with an assistant is that it wants you to be right. Accuse the gardener and the gardener becomes guilty; the clues you half-remember get retrofitted into a confession. Nothing was decided in advance, so nothing can contradict you.
All three Blackfile cases are written solution-first. The full answer and a random salt live in a reveal file whose SHA-256 hash is published before release. The card tells the assistant to fetch that file and referee against it, and every alibi, timetable and lie you are told comes out of a document that existed before you opened the chat.
When you finish, you read the reveal and paste it into /verify/ with the published hash. Your browser recomputes the digest. If it matches, the solution you just read is provably the one committed before you played — change a single character of it and the seal shows broken. The case cannot have moved to fit your guess, because a hash of a moved case is a different hash.
Questions about murder mysteries
How do I know the AI is not just making the answer up?
Because the answer was hashed before the game was published, and you check that hash yourself. After the reveal, paste the solution text and its salt into /verify/ along with the published commitment. Your browser recomputes SHA-256 locally and tells you whether it matches. A retrofitted solution produces a different digest, so the seal breaking is the tell.
Are the cases actually solvable, or do they need a leap?
Each case is checked cold before it ships: the culprit is uniquely determined by the evidence available to you, and every innocent suspect is excluded on at least two independent grounds. That check is part of the build, not a claim about intent — a case where two suspects both fit, or where the answer needs information you were never given, does not ship.
Can I replay a sealed mystery?
The solution is fixed, so a second run of the same case is a re-read rather than a new mystery — worth doing to see the clues you walked past, not to be surprised again. There are three Blackfile cases with unrelated solutions, and the social deduction and duel shelves use the same sealing method on setups that are hidden rather than deduced.