There I was, watching a glittering K-pop tableau sing about illegal substances with more swagger than a soft-drink ad. For three minutes the AI group “Obsession” made crystal meth, cocaine, cannabis and etomidate sound like quirky personalities at a party. Then the screen flipped to horror-movie discord—and the whole thing went viral for the wrong reasons.
I want to walk you through what happened, why it matters, and what this slip-up reveals about AI, design and public trust. You may think a public-safety spot is harmless; after you read this, you’ll see why tone and context are everything when persuasion meets synthetic media.
On International Day Against Drug Abuse, Hong Kong’s Correctional Services Department released an AI-made PSA that looked like a pop commercial.
The ad opens with a four-member virtual group—Obsession—each avatar pitched as a personified drug: “I’m Icy! Take a snort from me,” declares the character for crystal meth. The short plays like a glossy music video: bright hues, stylized dancers, bubble-pop hooks that register as fun before you reach the advisory.
That first half is the problem. The creative logic inverted the safety brief: glamour first, warning last. I’ve seen public-service messaging; when you sell the sensation before you name the risk, you invite curiosity, not caution. The ad’s upbeat choreography and productized lines made illegal substances read like lifestyle choices.
Will the ad make people want to try drugs?
Short answer: some viewers said yes. After the spot leaked and was reposted to YouTube, comments ranged from disgust to blunt curiosity—“I want to try a few bites of each,” one viewer wrote. Persuasion works by swapping attention for desire; when aesthetic pleasure comes before a sober warning, desire often wins.
Within hours people were split between outrage and amusement, and the CSD pulled the video after complaints.
The agency’s follow-up was clumsier: a second AI clip claimed—incorrectly—that taking or selling drugs wouldn’t send you to jail. That was posted and removed the same day, followed by a public apology on Facebook. The practical effect was a trust leak: an official body posted an ad that miscommunicated legal risk and glamorized illegal behavior.
I don’t blame AI tools—models don’t have judgment. But the humans who brief, edit and approve content do. When you hand creative control to synthetic imagery and slick audio without a hard editorial gate, you can create a message that wears the uniform of authority while conveying the opposite.
How did AI create such a misfire?
AI image and audio generators—think DALL·E, Midjourney, Stable Diffusion and audio tools—excel at surface appeal. They’ll produce a high-gloss band and catchy hook on demand. What they won’t do is understand moral framing or the ethics of tone. That’s why a public-service campaign that leans on AI needs clear human direction: strict scripts, legal vetting, and emotional choreography that keeps the warning in front.
On platforms like YouTube and Facebook, virality amplified the mismatch between form and message.
The ad’s aesthetic was stronger than its instruction; people shared it because it entertained, not because it educated. You can think of the spot as a Trojan horse for curiosity—slick entertainment that smuggled a dangerous idea past caution. Platforms reward engagement, and engagement doesn’t ask whether viewers learned the right lesson.
Platform mechanics matter. Algorithms surface glossy clips. Comments and reposts become evidence of impact. If you care about effect, you must design for more than clicks: you must design for comprehension, behavior and credibility.
Here’s what I’d advise anyone working with AI on public campaigns: name one single measure of success beyond views, lock scripts to legal facts before any generative work begins, and test with real audiences who can tell you whether the spot makes the hazard feel real instead of glamorous.
The Hong Kong case is a cautionary tale about tool fetishism and sloppy oversight. It also shows how easily aesthetics can hijack intent: a music-video polish can make a warning read like an invitation. The ad was pulled, apologized for, and discussed—so what now? Will agencies learn to treat generative AI as a powerful brush that still needs a steady human hand, or will more official messages arrive dressed as pop singles and tested only by like counts?