Definition
A type of AI model specifically designed to 'think through' problems step by step before giving an answer — producing a visible chain of reasoning rather than responding instantly. Examples include OpenAI's o-series and similar models from other labs. They tend to perform better on complex tasks but take longer and use more computing resources.
Why it matters
LRMs are being rapidly deployed for high-stakes decisions in legal, financial, and scientific domains because they appear more careful and trustworthy; however, their reasoning steps create a new attack surface (chain-of-thought hijacking) that standard models don't expose — meaning greater capability does not automatically mean greater security.