Check the two questions in the images. Looks fine, doesn't it?
The first one has 2 hidden characters.
The second one, just "Hello How are you", has 165.
You can't spot them. The AI still reads them.
Often, it's just copy-paste junk from an email or a web page.
But hidden text is also a known trick to sneak instructions into an AI (prompt injection).
So we flag these in our apps before you hit Send. One click to remove.
Small check. But in security, you miss one thing and all the other efforts are in vain.
How are you handling this, friends?