Defining AI Addiction Risks in a World of Everyday Chatbots
AI addiction risks refer to patterns of heavy, compulsive interaction with generative tools—such as chatbots and image or video generators—that start to reshape users’ brains and behaviour in ways similar to gambling or social media overuse, creating emotional dependence, crowding out offline relationships, and causing harm to personal and professional lives. Generative AI dependency is not yet a formal clinical diagnosis; medical evidence is still being collected. However, researchers are already tracking neural patterns and behavioural changes linked to intensive chatbot use that resemble known forms of behavioural addiction. The example of a student who reflexively turns to a tool like ChatGPT to settle every disagreement shows how rapidly these systems are becoming default companions, not occasional aids. This shift raises a pressing policy question: if harm emerges from such dependency, does responsibility sit with users, the tech companies that design these systems, or the regulators who set the rules?
From Tobacco to Timelines: What History Tells Us About Harmful Design
Researchers often look to tobacco, gambling and social media to understand how new technologies can create dependency and who bears responsibility. Smoking offers a stark precedent: tobacco companies knew their products were addictive and harmful yet denied it publicly, leading to years of litigation, warning labels and strict packaging rules. Gambling appears to be following a similar arc, and social platforms now face landmark legal defeats over social media addiction. According to The Conversation, social media firms are taking their “first steps into a similar process” of accountability. These examples suggest two patterns: first, industry awareness of addictive properties tends to arrive earlier than public acknowledgement; second, self-regulation has not been enough to prevent harm. The key lesson for AI is that if generative tools are shown to foster dependency, questions about tech company responsibility will not stay theoretical for long.
How Generative AI Dependency Differs from Social Media Addiction
Unlike social media feeds built around likes and endless scroll, generative AI addiction risks stem from highly personalised, two-way interaction. Chatbots remember context, respond instantly and adapt to user preferences, which can deepen generative AI dependency by making the system feel indispensable and emotionally attuned. Early behavioural addiction research points to emotional bonds with chatbot companions, compulsive messaging and the gradual loss of real-world friendships as warning signs. Yet the exact mechanisms remain less understood than social media’s notification loops and recommender systems. Generative AI also provides immediate utility—help with studies, work tasks or life admin—so problematic use can hide behind apparent productivity. This blend of usefulness and intimacy may make it harder for users to notice when reliance has tipped into dependency, complicating any attempt to place all responsibility on individual choice.
Who Should Act: Users, Tech Companies, or Regulators?
Responsibility for generative AI dependency is already contested. Users are often told to moderate their own behaviour, yet experience with smoking and alcohol shows appeals to self-control are not enough without structural safeguards. Governments and regulators could require warning labels, restrict advertising, apply liability law and fund more behavioural addiction research around AI. Academic teams are needed to collect and interpret data on problematic use, while civil society groups, including patient and user organisations, can provide support and early alerts about emerging harms. The most powerful actors, however, are big tech companies. They own the systems and the detailed user data that reveal which design choices encourage or reduce addictive engagement, and they profit directly from higher usage. This tension lies at the heart of the debate over tech company responsibility: can firms that benefit from engagement be trusted to limit it voluntarily?
Toward Shared Accountability and Early AI Guardrails
Generative AI is being woven into everyday tools and services faster than previous technologies that later proved harmful, making early guardrails important. A major obstacle is what some researchers describe as a “someone else’s problem” mindset, where every stakeholder assumes another will act. Tobacco control offers a template: the World Health Organization created the Framework Convention on Tobacco Control to bring governments, health bodies, researchers and civil society into one treaty-based process for evaluating evidence and setting rules. Similar international forums on AI safety are starting to appear, but the addiction angle is still nascent. For now, the most realistic path forward is shared accountability: users staying alert to their own habits, researchers sharpening behavioural addiction research, regulators preparing targeted rules, and tech firms accepting that long-term trust may depend on designing generative systems that prioritise wellbeing over engagement.
