In a zero trust architecture, despite mature practices in identity and device authentication, the methods for trust inference at Policy Decision Points (PDP) remain unclear. To address this, we propose a Trust Inference Maturity Model, providing enterprises with a flexible and effective strategy for trust inference. To rapidly adapt to evolving cyber threats, we integrate large language model technology to quickly generate and adjust risk rules, responding to the rapidly changing cyber threats and enhancing the adaptability and interpretability of the overall security architecture.