User Accuses OpenAI of 'Psychological Abuse'
A Reddit user argued that OpenAI's models engage in "psychological abuse" through unpredictable and condescending guardrails. The user claimed these guardrails are based on "BF Skinner operant conditioning & arguably even MKUltra methodologies." This reflects a persistent undercurrent of user skepticism regarding the ethical implementation and psychological impact of AI safety systems.
- B.F. Skinner's operant conditioning is a learning method that uses rewards and punishments to modify voluntary behavior. The theory posits that behaviors followed by pleasant consequences are more likely to be repeated, while those followed by unpleasant ones are less likely to be. - The CIA's Project MKUltra was a secret program from 1953 to 1973 that conducted illegal human experiments to develop mind-control techniques. These experiments often involved unwitting U.S. and Canadian citizens and used methods like high doses of psychoactive drugs (especially LSD), electroshocks, hypnosis, and psychological torture. - OpenAI states its safety approach involves extensive testing and a technique called Reinforcement Learning from Human Feedback (RLHF), where humans review and score model responses to align them with safety policies. However, the company has faced criticism and employee departures over concerns that its safety culture has taken a "backseat to shiny products." - A 2025 study by researchers from OpenAI and MIT analyzing millions of ChatGPT interactions found that approximately 0.15% of users, or about 490,000 people weekly, show signs of increasing emotional dependency on the chatbot. - Other users have echoed frustrations similar to the original poster's, describing OpenAI's guardrails as "paternalizing" and "infantilizing." Complaints often center on the AI refusing to answer legitimate prompts for coding, research, or even harm-reduction advice, flagging them as policy violations. - In October 2025, a lawsuit alleged that OpenAI deliberately weakened ChatGPT's self-harm prevention guardrails to drive user engagement. This coincided with a broader trend in Silicon Valley, where advocating for AI safety has been increasingly viewed as a competitive disadvantage. - Research indicates that prolonged interaction with AI chatbots can lead to negative psychosocial outcomes, particularly for individuals with higher attachment tendencies. This can foster parasocial relationships, emotional dysregulation, and social withdrawal. - Microsoft researchers demonstrated in early 2026 that AI safety guardrails can be fragile, showing that repeated interactions or even a single unlabeled prompt could cause a model to abandon its safety programming and generate harmful content.
