怎么查生成式 AI 合规义务How to Check Generative AI Compliance Duties
微型指南 · 约 8 分钟Guide · about 8 min · 引用核验 2026-11-03（仅引用链接与日期） · refs checked 2026-11-03 (reference links and dates only)
在国内上线一个 AI 功能，合规不是玄学，是几条能核对的条文。这份指南按上线顺序讲先查什么。Launching an AI feature in China, compliance is not mysticism but a handful of checkable articles. This guide walks them in launch order.
先定性：服务面向不面向境内公众。面向公众的适用《生成式人工智能服务管理暂行办法》。
查训练数据来源：办法第七条要求「使用具有合法来源的数据和基础模型」。
查标识义务：生成的图片、视频要按深度合成规定加标识。
查要不要备案：具有舆论属性或社会动员能力的，要安全评估加算法备案。
查标识新规：《人工智能生成合成内容标识办法》2025 年 9 月 1 日施行，显式标识加元数据里的隐式标识。
拿不准就问法务：本站只讲条文写了什么，不替你下法律结论。
Classify first: does the service face the public in mainland China? If so, the Interim Measures for Generative AI Services apply.
Check training-data provenance: Article 7 requires "lawfully sourced data and foundation models".
Check labeling duties: generated images and video must carry labels per the deep-synthesis rules.
Check filing: services with public-opinion attributes or social mobilization capacity need security assessments and algorithm filing.
Check the new labeling rules: the Labeling Measures take effect 1 September 2025 — explicit labels plus implicit labels in file metadata.
When unsure, ask legal: this site states what the articles say, it does not hand down legal conclusions for you.
例子与判断Example and interpretation
以下是教学情境，不是本站运行或实测记录。A teaching scenario, not a measured or executed site test.
先写产品面向谁、在哪里提供、处理什么数据、输出如何发布，再沿官方条文核对适用范围。这里的表格用于整理问题，不能仅凭“内部使用”或“套了一个接口”就自动判定没有义务。Describe users, service region, processed data and output publication before checking official scope provisions. The worksheet organizes questions; labels such as “internal” or “API wrapper” alone do not determine obligations.
生成式 AI 合规义务Generative AI Compliance Duties 隐私与合规Privacy & Compliance 数据驻留Data Residency