Cybersecurity of Frontier AI Models: A Regulatory Review — LessWrong