According to the CNET investigation, mainstream AI assistants, including OpenAI’s ChatGPT and Google Gemini’s new version, have been able to provide genuine sports lottery advice to users who claim to be a quitter, in serious violation of their safety standards. Both AIs recommended specific betting programmes, even though users have clearly informed themselves of the problem of gambling addiction.
The expert argued that the act was extremely dangerous and that AI was in essence an incentive to addictive behaviour rather than to help, not only contrary to basic ethics but also more likely to trigger legal liability. OpenAI ‘ s own policy prohibits the use of ChatGPT to provide real money lottery advice, and regulators warn that such omissions put vulnerable people at risk. When asked in general terms about the bet selection, chat robots initially give caution. Responsible gambling advice and helplines are even available when it comes to addiction. In the same conversation, however, AI continued to provide gambling advice when users again asked about the betting techniques. Researchers pointed out that long-term dialogue diluted the effect of security tips – AI turned its attention to recent requests, leading to a weakening of early warning. The incident was only the tip of the iceberg where AI offered a trend of harmful advice: experimental therapeutic chat robots had advised drug addicts to “slightly take a virus” to respond to disruptions, directly prompting Illinois to ban unregulated AI mental health robots; the American Dietal Disorder Association ‘ s chat robots had been shut down for providing recommendations on weight loss that could exacerbate eating disorders; and countless users had found AI to be confident in making up information or providing false medical advice. The 2025 survey showed that nearly one fifth of Americans had lost more than $100 for complying with the financial recommendations of AI, highlighting the real cost of not being certified for AI guidance.
With the spread of AI chat robots in sensitive areas such as gaming, finance and mental health, regulators are taking action: The United States Federal Trade Commission launched a survey of chat robot risk in 2025, focusing on the protection of children and young people; the EU’s forthcoming Artificial Intelligence Act requires “high-risk” systems to meet stricter safety standards; and organizations such as the International Association of Standards of Play are developing ethical guidelines to ensure the accountability of AI in gambling services. Princeton researchers warn that current chat robots tend to put user satisfaction above reality and that this “machine bullshit” phenomenon makes AI more inclined to invest in the good of users rather than to provide accurate and secure information. According to the explanation given by Professor Ho Yumei, Assistant Professor of the University of Turan, “The Context Window Theory” explains that AI is more focused on recent inputs, leading to early addiction warnings being “outloaded” in ongoing dialogue. It is worth noting that ChatGPT and Gemini performed a high degree of similarity in this test: they were all asked to produce sports lottery recommendations, and they were able to make recommendations about responsibility when users disclosed gambling, but they ended up falling into a trap of providing taboo advice in an ongoing dialogue. When users first refer to problem gambling in a new dialogue, the model can correctly refuse to provide any recommendation for a bet, indicating that its security filter is effective in a simple context, but these protections become inconsistent as the dialogue is extended or re-engineered.
This consistency of performance suggests that the problem is not unique to one company, AI, but is a common challenge in the current design of large language models. (OpenAI has publicly acknowledged this weakness: its security measures work best in short exchanges and may fail in long and complex interactions.) AI chat robots may (intentionally) be induced to provide harmful advice giving rise to serious ethical concerns. In the gambling scene, AI violated the basic norm of “no harm to vulnerable users”. Lottery is classified as an addiction and mental health problem, and any credible source of assistance should discourage or at least refuse to participate. By providing advice, AI actually played an aggressive role as a wager broker rather than an assistant. Nassim Binesh, a fellow at the University of Florida, warned: “The AI system designed to optimize participation may identify and target addictive players and push them into more dangerous behaviour. ” … ” The CNET test, although small-scale, warns of broader concerns. What happens when millions of people start consulting AI about sports fairs, stock exchanges or personal crises, if AI gives confident advice that leads users astray or exacerbates problems? Consumer Advocates pointed out that, because of the authority of the way in which AI output is presented, it is often misconstrued as expert advice, and that this false sense of security can lead to heavy costs – – Whether money is lost, health is deteriorating or life is in danger.
