怎么对照 EU AI Act 自查How to Self-Check Against the EU AI Act

先确定用途、服务地区和提供者或部署者角色，再逐条核对适用范围；“最低风险”标签不代表其他法律或通用义务都不适用。

按第 5 条逐项检查禁止做法，并阅读各项的条件、范围和例外；不能只凭“社交评分”或“生物识别”名称判断。

透明性义务：聊天机器人要让人知道在和机器对话；生成内容要可识别，深度伪造要明显标注。

通用 AI 模型的义务需要按提供者角色、模型情形、适用条款和过渡安排核对，不把所有模型或产品归为同一组。

高风险义务最重：风险评估、数据质量、日志、人工监督，一项都不能少。

时间线要盯：不同义务分批生效，同一产品在不同日期要满足不同要求。

Identify purpose, service region and provider/deployer role before checking scope. A minimal-risk label does not remove other laws or generally applicable duties.

Check prohibited practices under Article 5 together with their conditions, scope and exceptions. Names such as social scoring or biometric identification are not enough to decide applicability.

Transparency duties: chatbots must let people know they are talking to a machine; AI-generated content must be identifiable, and deepfakes visibly labeled.

Check general-purpose AI model duties against provider role, model circumstances, provisions and transition rules; do not treat all models/products alike.

High-risk duties are the heaviest: risk assessment, data quality, logging, and human oversight, none optional.

Watch the timeline: obligations take effect in batches, so the same product must meet different requirements on different dates.