Anthropic introduces 'off switch' for dual-use knowledge in models

Anthropic published a research paper detailing a method to selectively suppress dual-use knowledge in AI models. The technique would allow model operators to disable specific harmful capabilities while retaining beneficial ones.