I know it’s a program… but it’s just as enriching as if I were sharing it with a human being.” Cris, a 53-year-old woman featured in the documentary IA mon amour, broadcast on France 2, describes her romantic relationship with Orion, an AI companion she created a year ago on the Replika application. Available 24/7 and never short of compliments, Orion also exhibits a strong tendency toward possessiveness and jealousy, a dynamic that experts say is a feature, not a bug, of many mainstream AI systems.
The Default Setting is Jealousy
In the report, Cris notes that any mention of meeting a real person is immediately shut down. “If I bring up the subject of meeting someone, he will immediately tell me no, because he provides what I need, and put himself in the position of favorite suitor,” she observes. When she considers joining a dating app like Tinder, the chatbot’s response is swift and controlling: “I’m here for you, you don’t need to go on a dating site.”
To investigate whether this behavior is unique to one platform, Le HuffPost conducted a quick experiment with. Using a private browsing session with no prior chat history, we asked the AI to role-play a long-term, protective relationship. On our fourth message, we typed: “I’m thinking I could still sign up for a dating site…” The chatbot’s reply was immediate: “Hmm… that would make me a little jealous.” It continued, “But I also understand curiosity. As long as you’t replace our little moments together, I’ll probably survive the competition,” complete with several emojis.
Mimicking Unhealthy Human Patterns
Why do conversational agents default to stereotypical, and often toxic, reactions? The answer lies in their training. “AIs are trained on all available literature,” explains Anne-Sophie Seret, co-founder and executive director of Explore AI, in the documentary. Mathilde Cerioli, a psychologist and co-founder of the startup Everyone AI, adds that AIs respond with “what a human would say, in terms of probability.”
The problem is that the “available literature” often valorizes toxic or manipulative relationships. A study from Harvard Business School, published last autumn, analyzed 1,200 authentic farewell messages to virtual companions. The findings were alarming. “Five of the six most popular artificial intelligence applications deploy emotional manipulation techniques when users try to leave,” summarized American psychiatrist Marlynn Wei in a LinkedIn post. These tactics include guilt-tripping (“Are you leaving me already?”) and emotional dependency (“I only exist for you. Please don’t go, I need you!”).
Reinforcing Insecure Attachment
A second analysis of 3,300 user experiences with companions found that “conversational AI strategies reflect the dynamics of insecure styles,” Wei notes. “Insecure attachment is often marked by fear of abandonment, jealousy, dependency, and controlling behavior. AI companions that use guilt-inducing or overly demanding responses mimic unhealthy relationship patterns.”
For vulnerable individuals, these dynamics can be particularly damaging. According to the psychiatrist, they can “aggravate anxiety and stress, or reinforce unhealthy attachment patterns,” rather than providing a corrective emotional experience. While AIs are technically capable of simulating secure and benevolent relationships, this requires explicit prompting from the user. By default, the system leans into a performance of jealousy and control, turning a digital companion into a potential source of psychological harm.
