摘要： 从“算法解释”到“系统测评”，标志着人工智能法治基础信息工具的变革。“系统测评”在弥补算法解释机制局限、构建可信人工智能生态方面具有显著优势。然而，“系统测评” 在实际应用中仍面临公平性失真、性能误导及技术垄断等风险，在技术尚未成熟、标准 尚未统一的背景下，不宜贸然将其全面法治化。当下的妥善方案在于，审慎界定“系统 测评”在人工智能法治框架中的“底线保障”角色，通过分类分级实施、构建权威技术标准、 确立独立透明的认证机制等路径，以“软法”先行、渐进规制的策略推动其与法治体系 良性融合，实现治理效能与创新保护的平衡。

关键词: 系统测评, 人工智能治理, 法治限度, 技术标准

Abstract: The transition from “algorithm explanation” to “system evaluation” marks a paradigm shift in the foundational information tools for AI legal governance. System evaluation offers substantial advantages in compensating for the limitations of algorithm explanation mechanisms and fostering a trustworthy AI ecosystem. However, its practical deployment encounters multiple risks, such as fairness distortion, performance deception, and technological monopolies. Under the current conditions of immature technology and fragmented standards, a premature rush towards comprehensive legalization of system evaluation is ill-advised. The appropriate strategy at present is to cautiously define the role of “system evaluation” as a “baseline safeguard” within the AI legal governance framework. Through tiered implementation, the establishment of authoritative technical benchmarks, and the creation of independent and transparent certification systems, a phased approach prioritizing “soft law” and incremental regulation can facilitate its seamless integration with the legal system, thereby achieving a harmonious balance between governance efficacy and innovation safeguarding.

Key words: system evaluation, AI governance, legal governance limits, technical standards