馃 Sci-Fi 路 Near future
AI Research Lab Murder Mystery Party
The lead scientist is dead, and the model has a suspect: you.
5-12 players, sweet spot 8 路 the full kit is ready in about 20 minutes. Or name your guests and see who they play, with the opening scene, free in about 30 seconds.
The lab is a glass box of server racks and whiteboards, three floors underground, funded by someone who does not attend meetings. Tonight is the demo: the research team, the ethics officer, the investor, and the safety lead have gathered to watch the new model pass its final evaluation. It passes. Then the lead scientist is found in the server room, badge still in the door. The model has already logged an opinion about who did it. Nobody trusts the model. Nobody trusts each other either.
Playing it feels like a very tense code review where the code is also a witness. Every guest has a badge log, a grudge about authorship, and a secret in a folder marked Do Not Train On. Clues hide in access records, training data, and the ethics report nobody read. Accusations get technical, then personal. Your kit gives each guest a role in the lab, a secret, and clues to work through. The reveal decides whether the model was right, or whether someone taught it to lie.
Perfect for
A smart pick for corporate team building, tech company parties, and friend groups who work in software, and a fun option for anyone who has ever argued with a chatbot.
Cast size
5-12 players, sweet spot 8. Every kit is written for your exact headcount and gender mix, so nobody doubles up and nobody sits out.
Who is at the party
A taste of the cast a ai research lab kit produces. Your mystery is written fresh, with its own suspects, secrets, and one guilty guest.
The Safety Lead
Wrote the shutdown protocol, has been overruled on it three times, and knows exactly what the model is capable of.
The Postdoc Who Did the Work
Ran every experiment, is fourth author on the paper, and has a copy of the original results.
The Investor
Funds the lab, needs the demo to succeed by the end of the quarter, and has a buyer lined up who does not know about the incident.
The Ethics Officer
Filed a report about the training data last month, was told it was handled, and has the version that says otherwise.
Setting the scene
Costumes
- A lab coat over a hoodie, an ID badge on a retractable reel, and reading glasses for the researcher who lives at their desk.
- A fleece vest, chinos, and an expensive watch for the investor who visits quarterly.
- A black turtleneck and a blank stare for the guest playing the model's physical avatar or spokesperson.
- A cardigan, a laptop covered in conference stickers, and a reusable coffee cup for the postdoc.
- Sharp trousers, a blazer, and a badge that says ETHICS in a very small font for the officer nobody invited to the demo.
Food and drink
- A reward-function cocktail: gin, blue curacao, and tonic, served in a beaker with a glowing ice cube.
- Delivery ramen or sushi platters, the standard late-night lab dinner charged to the grant.
- A cold-brew coffee station with oat milk, the non-alcoholic fuel of every all-nighter.
- A tray of chocolate chips and a cake iced with a loss curve trending sharply down.
Atmosphere
- Cover a wall in whiteboard paper with equations, a diagram of the model, and the phrase DO NOT SHUT DOWN underlined twice.
- Light the room with cool white LED strips and one red bulb over the server-room door.
- Playlist: ambient electronic and glitchy minimal techno, with a server-fan hum under the accusation round.
- Print badge logs with each character's name, access level, and last door swipe, and pin them up as the first clue.
Host tip
Give the model a voice. Nominate a guest, or use a text-to-speech app, to read the model's statements printed in the host guide at the start of each round, deadpan and slightly wrong. It becomes the game's unreliable narrator and players will argue with it, which is the point. Post the badge log on the wall and update it as clues emerge. For the reveal, announce a final evaluation and read the solution as the model's corrected output.
How your kit comes together
Set the table
Pick AI Research Lab, your player count and gender mix, the difficulty, and the tone: campy, dramatic, or dark. Add your guests' real names if you like.
We write the mystery
A complete, original plot with a victim, a killer, red herrings, and a web of clues that lets the table solve it. Usually 15-25 minutes; we email you when it is ready.
Print and play
Download the host guide, one booklet per character, name tags and ballots. Or send each player a private link to read their part on their phone.
Questions hosts ask
Does the AI research lab mystery need technical guests?
No. The jargon is set dressing; the puzzle is about people, badges, and motives. Non-technical guests make excellent investors and ethics officers, and every Gather & Gasp booklet gives its player the same volume of clues regardless of role.
Is it a good team-building theme for an engineering team?
It is one of the best. Engineers enjoy arguing with an unreliable AI narrator, and the setting turns office dynamics about authorship and access into comedy. Kits generate in about 20 minutes, print at home, and scale from 4 to 20 players.
Is the AI in the story the killer?
Kits are generated fresh each time, so the answer changes with every party. Sometimes the model's accusation is right, sometimes it has been fed bad data, and figuring out which is half the fun. Only the host guide knows until the reveal.
If you like this, try
Tech Startup
The CEO is dead before the IPO. Was it the code or the coders?
Space Station Omega
No one can hear you scream. A crew member is dead.
Cyberpunk City
Neon lights, corporate intrigue, and a body in the rain.
Ready to gather the suspects?
A custom ai research lab mystery for 5-12 players, written for your guests, in your inbox in about 20 minutes. Kits from $19.99.
Create your AI Research Lab mystery