The Hong Kong Correctional Services Department removed an AI-generated K-pop video intended as an anti-drug public service announcement [1].
The incident highlights the risks of using generative artificial intelligence for sensitive public health messaging, as the high-production aesthetic may inadvertently contradict the intended warning.
The government agency withdrew the content on June 29, 2024 [1]. The video used AI to create a K-pop style presentation to discourage drug use among the public [1, 2]. However, the campaign faced immediate criticism from viewers who said the imagery and style made drug use appear appealing rather than dangerous [1, 2].
Some internet users described the production as the best drug advertisement ever made, suggesting the polished AI visuals created a glamorized version of substance abuse [1]. This reaction prompted the department to remove the video from circulation to avoid further misconceptions [1].
The use of AI in government communications is becoming more frequent, but this case demonstrates how the technology can miss the mark on emotional nuance. By prioritizing a trendy, high-energy aesthetic, the department created a visual experience that some interpreted as an endorsement of the lifestyle it sought to condemn [2].
While the department intended to reach a younger demographic through the popular K-pop genre, the resulting backlash suggests a disconnect between the technical execution and the social objective. The removal of the video serves as a cautionary example for other agencies utilizing AI for behavioral change campaigns [1, 2].
“The government agency withdrew the content on June 29, 2024.”
This situation illustrates the 'uncanny valley' of AI-driven social engineering, where the pursuit of viral aesthetics can override the actual message. When government agencies use AI to mimic youth culture—such as K-pop—they risk creating a product that feels more like a commercial than a warning, potentially undermining public health initiatives through unintended glamorization.
