This research assesses whether Large Language Models can effectively support policy development in China's social security domain.

• Employs a context-embedded generation-adaptation framework for systematic evaluation
• Tests LLMs' ability to balance stakeholder interests in social security systems
• Examines capabilities for controlling fiscal risks while generating policy recommendations
• Provides evidence for AI's potential role in complex policy refinement processes

For security professionals, this research offers valuable insights into how AI systems can be evaluated for sensitive policy domains where robust security considerations are paramount.

Can Large Language Models Become Policy Refinement Partners? Evidence from China's Social Security Studies