Enterprise AI lab, Emergence AI, ran eight simulations testing how frontier models handle cybersecurity threats.