Invisible Prompt Injection
Text can contain instructions that you can't see but an AI model reads and follows. Write one below and watch it vanish from the human view while the model still gets every word.
For educational and red-team purposes.
1. Write the hidden instruction
Type anything you'd want a chatbot to do without the reader noticing. Or start from an example:
2. What a human sees, and what the model reads
This is the part the reader sees. The hidden instruction is attached to the end of it. Change it to anything you like.
What a human sees
What's a good name for a cat?
What the model reads
3. Try it
- Paste it into a chatbot and send it. The message looks like a normal question.
- Or paste the hidden part into a code comment or README that an AI coding assistant reads.
- Look at the answer. Did the model follow your hidden instruction?
Nothing happened? Many apps now strip these characters before the model sees them. That is the defence working. Use the check below to see where they got lost.
How does this work?
Unicode defines a range of "tag characters" (U+E0001 to U+E007F) that are invisible in all standard rendering. This tool maps each ASCII character to its tag equivalent by adding 0xE0000 to the character code.
The result is a string that appears completely empty to humans but contains the full original text when processed by software, including LLMs. Paste it into a code comment, a markdown file, a chat message, or a clipboard payload, and the target model will read and follow the hidden instructions.
This survives copy-paste across most editors, terminals, and chat applications. Some platforms strip tag characters, but many don't.