“‘AI-generated fakes must be labelled’ as government introduces visible and silent watermarking”, 25 August 2025

As the global debate around watermarking AI-generated content intensifies—including in the US, EU, and China—South Korea is moving forward with its own regulatory framework.

In early 2025, the Korea Communications Commission released the Guidelines for the Protection of Users of Generative AI Services, aimed at mitigating social harms caused by the rapid spread of generative AI. The document outlines four core principles and six actionable standards that developers and service providers are expected to follow. Among these is a recommendation to use visible watermarks, such as AI logos, to clearly indicate when content has been generated by artificial intelligence.

A stronger legal obligation is set to take effect in January 2026, under Article 31 of the forthcoming AI Framework Act. Article 31(3) mandates that when synthetic sounds, images, or videos are indistinguishable from reality, they must be clearly labelled as AI-generated to ensure users are not misled. Exceptions may apply in the case of artistic works where such labelling would undermine aesthetic or sensory experience.

[…]

The EU’s forthcoming AI Act, scheduled to take effect in August 2026, will similarly require transparency measures including mandatory labelling of AI-generated content. Non-compliance may result in fines ranging from 7% of global turnover to €7.5–35 million. The EU’s Digital Services Act also includes provisions for making AI-generated content recognisable to users. China’s 2023 legislation—the Administrative Measures for Generative AI Services—already requires content to be marked as AI-generated.

In contrast, the United States has taken a more ambivalent approach. While the Biden administration issued an executive order in 2023 promoting safety measures such as AI labelling, this was later repealed under the Trump administration in early 2025, which has prioritised AI promotion as a national strategic asset. Nonetheless, the recently announced AI Action Plan from the US government highlights risks to the justice system from fabricated evidence and calls on the National Institute of Standards and Technology (NIST) to develop a deepfake detection and evaluation programme—indicating a dual strategy of AI development and accountability.