Study finds watching AI's thoughts may reveal dangerous intent - Tech in Asia

Researchers suggest that tracking the step-by-step reasoning—called “chain of thought”—in AI models could help detect harmful behavior early, ...