They hide inside documents, and we can’t see them.
Someone writes a document. Inside it, they put a sentence in white on white. To us, it’s invisible. The AI reads it anyway. And the sentence tells the AI: side with this document. It has already happened in the US, in a lawsuit against a clinic.
What to do right away: before letting AI read a document, let’s copy it into plain text: the hidden sentence pops right out. And the final decision is always made by a person.
So say it at the office: every document read by AI gets read again by a person!
