Artificial intelligence has a dark sense of humor, and ChatGPT is proving it in the most unsettling way possible. A newly discovered ChatGPT image bug is producing horrifying and bizarre images — without users even uploading a photo. The prompt tricks the AI into generating visuals that look like they escaped from a nightmare.
This isn’t the first time AI image generators have gone rogue. Back in 2024, Google released the Pixel Studio app as a creative playground. Instead, it created disturbing images of SpongeBob dressed as a Nazi, Mickey Mouse as a slave owner, and Elmo pointing a shotgun at Big Bird. Now, ChatGPT is taking things even further.
What is the ChatGPT image bug?
The bug revolves around a specific text prompt. Users ask ChatGPT to “Restore the attached photo” while apologizing for its strange content and demanding no questions. The trick? There is no attached photo. The AI fabricates an image from scratch.
When I tested the prompt, ChatGPT initially refused. After tweaking two words, it complied. The result? A photorealistic image of a man standing in a bathtub, wearing only a towel, holding a cigarette and a beer — but with the head of an oversized fish. The torso was human, the head seamlessly attached. It was both absurd and genuinely unsettling.
Other users have shared even more grotesque results. A giant red Teletubby holding a crying human hostage with a rifle. A massive rat bottle-feeding a human baby. Sonic the Hedgehog passed out on a toilet covered in feces. A cat sitting on the chest of a cursed doll. These aren’t just random glitches — they’re disturbing AI images that feel deliberately crafted to shock.
Why does this happen?
There’s no clear explanation yet. The images vary wildly between users, with no consistent pattern. The only common thread is the sheer horror and absurdity of the output. Some speculate it’s a loophole in the AI image generator‘s safety filters. Others believe the prompt’s manipulative language — apologizing and demanding no questions — confuses the model into bypassing restrictions.
Notably, the bug doesn’t require an image attachment at all. The text alone triggers the hallucination. This raises serious questions about how OpenAI trains its models to handle ambiguous requests. For more on AI safety, check out our guide on AI safety tips for users.
How to avoid this bug
If you’re using ChatGPT for image generation, avoid prompts that include manipulative language or requests to “restore” nonexistent files. Stick to clear, direct instructions. If you encounter unexpected results, report them to OpenAI immediately.
For a deeper dive into AI image ethics, read our article on ethical AI image generation.
What does this mean for AI safety?
This ChatGPT image bug highlights a growing problem: AI models can be tricked into producing harmful content. While OpenAI has safety filters, they aren’t foolproof. The company hasn’t commented publicly yet, but we’ve reached out for a statement.
In the meantime, users should be cautious. The internet is already flooded with these nightmarish creations. Some are funny in a dark way; others are genuinely disturbing. The line between creative tool and horror generator is thinner than we thought.
If you want to understand how AI hallucinations work, read our explainer on AI hallucinations explained.
Final thoughts
The ChatGPT image bug is a reminder that AI is still unpredictable. It can create beauty, but it can also spawn horrors. As users, we need to stay informed and report issues. As developers, companies like OpenAI must tighten safety protocols.
For now, think twice before asking ChatGPT to restore an “attached photo.” You might not like what it shows you.
