跳到主要内容Skip to main content

← 返回链← Back to Chain

来源标识Provenance

C2PA 面向 AI/ML 的实施指南C2PA guidance for AI and machine learning

交集:把内容凭证从「媒体文件」扩到 AI 供应链的三件东西——模型凭证、训练数据凭证、输出凭证;输出用 trainedAlgorithmicMedia 或 trainedAlgorithmicData 标记来源类型,模型还可挂一条数据挖掘断言,声明是否允许被拿去训练别的模型。适用范围:规范给的是断言类型与清单结构,用于发现篡改与记录来源,不评估模型质量或输出是否正确。未解决:一切都依赖提供方如实填写,签名只覆盖签名之后的改动;凭证能否在转码与二次分发后存活取决于各平台。该文档本身以 CC BY 4.0 发布,可转载但需署名。The intersection: extending content credentials from media files to three things in the AI supply chain — a model credential, a training-data credential and an output credential. Outputs are tagged with trainedAlgorithmicMedia or trainedAlgorithmicData as the source type, and a model may carry a data mining assertion stating whether reuse for training other models is allowed, constrained or not allowed. Scope: the guidance supplies assertion types and manifest structure for detecting tampering and recording origin; it does not assess model quality or whether an output is correct. Open: everything depends on the provider recording honestly, a signature only covers changes made after signing, and whether a credential survives transcoding and redistribution depends on each platform. The document itself is published under CC BY 4.0, so reuse requires attribution.

关键参数Key specs

覆盖对象 / scope
模型凭证 · 训练数据凭证 · 输出凭证
输出来源类型 / source type
trainedAlgorithmicMedia(媒体)· trainedAlgorithmicData(非媒体)
权利表达 / rights
data mining assertion:allowed / constrained / not allowed
文档许可 / licence
CC BY 4.0

官方原文怎么说What the source says

The C2PA specification can be used to add cryptographic information to detect the tampering of media files and streams.
1 https://spec.c2pa.org/specifications/specifications/1.4/ai-ml/ai_ml.html
The Content Credential for an AI-ML model provides the consumer, e.g. a system operator designing an AI-ML system, with provenance and authenticity of the model.
2 https://spec.c2pa.org/specifications/specifications/1.4/ai-ml/ai_ml.html
the model C2PA Manifest can contain a data mining assertion that asserts rights as allowed, constrained, or not allowed for generative or non-generative AI-ML models or both.
3 https://spec.c2pa.org/specifications/specifications/1.4/ai-ml/ai_ml.html

资料与许可Source and licence

资料时间Source checked
2026-11-07
许可 / 类型Licence / kind
official-specification
许可说明Licence evidence
https://spec.c2pa.org/specifications/specifications/1.4/ai-ml/ai_ml.html
类目Category
来源标识Provenance

去官网看原页Open the official page →

← 返回链← Back to Chain