UK institutions call for strengthening frontier AI safety management

The UK AI Safety Institute published an article on the 12th stating that as frontier AI capabilities rapidly advance, security risks are shifting forward to the model development stage, and existing safety assessment methods are struggling to keep pace with technological development. It is necessary to explore new technical safety practices to ensure that frontier AI development activities proceed safely. The article pointed out that previously, AI model safety assessments mainly focused on ensuring safety at the time of public release, but today some AI systems have enhanced autonomous capabilities, especially in the field of cybersecurity. Even if a model has not yet been publicly released, it may pose risks during training, internal deployment, and evaluation.