Anthropic is working on a 'switch button' to make AI forget its most dangerous skills

Anthropic has unveiled GRAM, an experimental AI safety system that could let models "forget" dangerous knowledge like hacking ...