“Yoshua Bengio states that over the past year, as frontier AI models have improved their reasoning capabilities, they have exhibited more misaligned and undesirable behavior.”