X @Anthropic

The reason many LLMs don't fake alignment isn't lack of ability. Base models (which don’t have training to be helpful, honest, and harmless) sometimes fake alignment, suggesting they have the underlying skills. ...