UK AI Security Institute sets eight priorities for voluntary AI development safeguards

The institute says risks now arise during training and internal use, while evaluation methods struggle to keep up. Mature practices could eventually become international technical standards.