More than 1,500 AI projects are now vulnerable to a silent exploit

Sounds interesting... looking into research paper, it's much more than just indirect prompt injection