OpenAI is building an artificial-intelligence system dubbed “Universal Verifier” that automatically assesses the answers generated by its large language models, The Information reported on Monday. The tool is intended to score responses not only in well-defined tasks such as coding and mathematics but also in more subjective areas including business decision-making and creative writing. By acting as an internal reviewer, the verifier provides reinforcement-learning feedback that could reduce the company’s reliance on human annotators and accelerate model improvements. People familiar with the project said the technology is being integrated into the training pipeline for GPT-5, the planned successor to GPT-4. The work extends research described in the paper “Prover-Verifier Games Improve Legibility of LLMs,” which details a production-ready pipeline in which a dedicated verifier model evaluates each reasoning chain before an answer is accepted.