CAC finalizes AI labeling rules
In September, we told you that China's cyberspace regulator (CAC) had released new draft rules requiring generative AI tools to clearly label AI-generated content (AIGC).
- The rules specify what type of metadata and watermarking should be added to text, audio, video, and other machine-made media to ensure it can be traced back to the platform where it was generated.
- They also list the responsibilities of tech platforms and app stores to ensure AIGC is labeled.
On Friday, the CAC and three other ministries jointly released the finalized version of the rules.
- Very little has changed from the original draft.
- Companies must be compliant by September 1, when the rules take effect.
Some context: This is China's fourth AI-related regulation.
- Previous regulations govern recommendation algorithms, deepfakes, and AIGC more broadly.
More context: Several of those existing regulations already require AIGC content to be labeled as such.
- But these new regulations flesh out more specific details on content labeling requirements.
A new mandatory technical standard on AIGC content labeling was released in coordination with the regulation to help AI companies comply.
Our take: The simultaneous release of a regulation and a matching standard is particularly notable.
- That rarely happens, and it indicates a tighter coordination between policymakers and techies.
Our other take: Chinese AI regulations are all but guaranteed to place more importance on generated content traceability than regs in other countries.
- That's because regulators are deeply concerned about the possibility that AIGC will violate censorship rules, and are thus tying AIGC into China's broader content tracing and control ecosystem.