“Even the most advanced frontier AI models currently have a high rate of false positives.”