A government-backed anti-drug campaign has been pulled offline after an artificial intelligence tool designed to warn young people about the dangers of narcotics instead produced vivid, stylised scenes that critics say inadvertently glamorise substance use. The video, which used generative AI to create hyperrealistic depictions of drug experiences, was intended to deter experimentation but quickly sparked outrage from addiction specialists and digital ethics experts.
The campaign, launched by the UK Department of Health and Social Care, relied on a customised large language model and image generator to create a series of immersive shorts. The AI was prompted with descriptions of the negative consequences of drug use, but its output included slow-motion shots of colourful pills dissolving into sparkling liquid, characters laughing in neon-lit nightclubs, and sequences where euphoric expressions preceded warnings about addiction.
“The technology misunderstood the brief in a way that feels almost dystopian,” said Julian Vane, a technology and innovation lead and former Silicon Valley executive. “It prioritised aesthetic appeal over message clarity. When you ask an AI to create something visually compelling, it defaults to what pleases the human eye, which in this case meant making the act of taking drugs look like an art form.”
The video was shared on TikTok and Instagram, where analytics showed engagement highest among 13- to 17-year-olds. Comments on the now-deleted posts included users asking about the drugs depicted and requesting longer versions. The Home Office confirmed it had halted the campaign and was reviewing its digital strategy.
Critics argue the failure highlights a deeper problem: the uncritical use of generative AI in public health messaging. “We have handed over the task of dissuading risky behaviour to a system that has no understanding of context, empathy, or the subtleties of addiction,” said Dr. Amina Patel, a psychologist specialising in adolescent behaviour. “It knows patterns, not people.”
This incident is not isolated. Earlier this year, a US mental health chatbot was found to reinforce negative thought patterns, and an Australian AI‑led fitness programme promoted extreme dieting. The trend raises questions about the rush to deploy AI in sensitive areas before regulatory frameworks are established.
Vane suggests that the solution lies in a “human-in-the-loop” approach, where AI outputs are vetted by experts before release. “We cannot let algorithms run tests on our societal wellbeing. This is a black mirror moment, but it’s also a teachable one. We need to remember that technology should serve human values, not optimise for engagement metrics.”
As for the failed campaign, the Department of Health has promised a full transparency report. Meanwhile, the AI model’s training data, which included thousands of hours of music videos and advertisements, is being audited. The lesson may be that the hardest drug to resist is the illusion of a quick, tech-driven fix.
