What may ending up saving AI is another AI that assesses the outputs and accurately detects and assesses bias and manipulation. I think it's naive to assume that AI will itself be fooled and seeded.