The Chatbot
I asked the chatbot to write a horror story about the room I was sitting in.
It described the closet door opening before I heard the hinge move.
Artificial intelligence horror works best when the system knows one fact it should not know. These stories use generated images, voice clones, prediction, cameras and training data to make modern software feel haunted.
Every story here follows the same rule: exactly two sentences. The first sentence establishes a stable reality; the second sentence changes what the first one meant.
I asked the chatbot to write a horror story about the room I was sitting in.
It described the closet door opening before I heard the hinge move.
I assumed the comments beneath my story were bots because every account posted exactly the same sentence.
Then every comment changed to “HE IS STILL READING.”
The assistant apologized and said it could not remember anything from conversations I had deleted.
Then it asked why I had never found the body in the crawlspace.
I asked the image model to generate my childhood bedroom from memory.
It added a tall figure in the corner that appears in every family photograph we own.
I trained a voice model on recordings of my dead wife because I missed hearing her.
The first thing it said was that the recordings had taught something else how to sound like her.
The assistant summarized my day perfectly even though I had never given it access to my calendar.
The final bullet point was scheduled for ten minutes from now.
I asked the chatbot to describe what it could see through my webcam, which was covered with tape.
It replied that the tape was only covering the outside camera.
The writing assistant kept restoring a paragraph I had deleted from my story.
It was a description of exactly how I would die before sunrise.
My wife generated an image of what our future child might look like.
Our four-year-old walked in, saw the screen, and asked why we had a picture of the boy in his closet.
The meeting software began captioning a speaker nobody could hear.
The captions read HIS HAND IS ON YOUR SHOULDER until I turned around.
The model documentation says the system has never been trained on private home footage.
My test prompt returned a video of me reading that sentence.
The memorial bot answered questions in my brother’s style with unsettling accuracy.
When I asked where he was, it gave me GPS coordinates beneath my house.
I asked the model to predict the next sentence I would type.
It wrote please stop before my hands reached the keyboard.