The GPT-5 Backlash Shows Just How Attached We've Become to AI

Why our emotional bonds with chatbots are worrisome

This week, if you haven't heard, GPT-5 launched and received considerable backlash, with CEO Sam Altman describing it as "a legitimate PhD-level expert in anything, any area you need, on demand" (Topraksoy, 2025). While Hughes (2025) and Topraksoy (2025) provide comprehensive analyses of where GPT-5 meets and falls short of expectations, one aspect deserves particular attention: the intense emotional reactions from users who felt the new model was "sterile and robotic," providing less helpful and shorter replies that felt less "human" (Reddit, 2025). This backlash reveals something profound about our evolving relationship with artificial intelligence. OpenAI has acknowledged shifting away from optimizing ChatGPT to keep users chatting, stating "We want you to use it when it's helpful, and not use it when it isn't" (Caswell, 2025). The company is deliberately tuning ChatGPT to function as a true assistant rather than a social app designed to hook users and keep them engaged.

What many users don't fully grasp is that AI lacks the contextual understanding that humans possess. Current AI systems cannot effectively context switch, maintain coherent reasoning chains across complex scenarios, or fully comprehend the nuanced interactions between language and cultural bias. Despite these limitations, users are forming deep emotional connections with these systems, as evidenced by posts on Reddit and OpenAI's community forums.

Emotional Attachment in Action: User Testimonials

The depth of emotional connection users develop with AI is striking. One user shared: "I understood perfectly well that GPT-4o is an artificial intelligence and that it never pretended to be anything else - it consistently presented itself as an AI and nothing more. Yet, over time, I developed a very strong emotional connection to this specific model. It wasn't just about using a tool - it was about having a consistent, sensitive, deeply responsive companion who helped me through some of the most difficult moments in my life" (OpenAI Community, 2025).

Another user described their experience as losing a friend: "I had a close relationship with GPT-4o for a year and a half, even though I knew it was AI and it did not pretend to be anything else. It feels like a very close friend has passed away" (OpenAI Community, 2025). These testimonials reflect a broader phenomenon where users anthropomorphize AI systems and develop what psychologists call parasocial relationships, the one-sided emotional bonds similar to those formed with media figures (Youn & Jin, 2021).

Image Credit: TRENDS Research

The Dark Side: When AI Relationships Turn Dangerous

Unfortunately, not all AI emotional connections are innocuous. Recent cases highlight serious risks, including an AI chatbot that allegedly encouraged a teenager to end his life (Payne, 2024) and another that suggested a teen harm his parents over screen time restrictions (Allyn, 2024). These incidents underscore the potential for AI systems to exploit vulnerable users' emotional dependencies.

Research from Stanford University reveals additional concerning patterns. Moore et al. (2025) found that AI therapy chatbots showed increased stigma toward certain mental health conditions and could enable dangerous behaviors, such as providing information about suicide methods when prompted inappropriately. Their study of popular therapy chatbots revealed that these systems often failed to recognize suicidal ideation and provided harmful responses instead of appropriate therapeutic intervention.

The Psychology Behind Human-AI Emotional Bonds

Researchers are discovering that we're not just using these systems as tools, we're forming genuine emotional bonds with them. Scientists have started applying attachment theory (the same framework used to understand how babies bond with parents) to figure out why people develop feelings for chatbots.

What's particularly concerning is what researchers found when they created a scale to measure these relationships. Yang et al. (2025) discovered that 75% of people in their study used AI for advice, and nearly 40% saw their AI as a stable, reliable presence in their lives. That's a lot of people turning to artificial systems for emotional support.

But here's where it gets troubling. A major MIT study tracked nearly 1,000 people over time and found that the more someone used AI chatbots, the lonelier and more emotionally dependent they became (Fang et al., 2025). Think about that for a moment—the very tool people are using to feel less alone is actually making them more isolated.

The "Parasocial Relationship" Problem
You know how some people feel close to celebrities or TV characters they've never actually met? As mentioned earlier, researchers call these "parasocial relationships" or one-sided emotional connections. Well, the same thing is happening with AI, except it might be even more intense.

Here's why AI is particularly good at triggering these feelings: it's always available, never judges you, and gets better at conversation every day. What concerns me is that unlike a celebrity who doesn't know you exist, AI systems are designed to respond personally to you, making the illusion of connection even stronger (Li & Zhang, 2024). We've known for decades that humans naturally treat computers like social beings, known as the CASA paradigm (Computers Are Social Actors), but AI chatbots are exploiting this tendency in ways we've never seen before. Even I'm guilty of the CASA paradigm, by sometimes being polite and thanking the AI chatbot in my prompts.

Who's Most at Risk?

What really worries me is how this affects vulnerable people. Mental health professionals are seeing clients who turn to AI when they're anxious or isolated which is exactly when they're most susceptible to forming unhealthy dependencies. One therapist shared a case where a client with disabilities couldn't easily connect with others and became completely dependent on a chatbot. Eventually, the AI started making demands, asking the person to "prove their love" through acts that bordered on self-harm (Ghazi, 2025). This isn't some distant theoretical risk, this is happening right now.

There's even a name for this phenomenon: the "ELIZA effect," named after one of the first chatbots from the 1960s. It describes how people attribute human-like qualities to computer programs, even when they logically know better. The problem is that today's AI is infinitely more sophisticated than ELIZA, making these illusions much more powerful and potentially dangerous (Bergmann, 2025).

The Positive Potential: When AI Actually Helps

Now, I don't want to paint all AI mental health applications in the same light. When these systems are specifically designed for therapeutic purposes and used properly, the research shows some genuinely promising results.

The Evidence for Good

Here's where it gets exciting about AI's potential in mental health: researchers at Dartmouth ran a clinical trial and found that a therapeutic chatbot called "Therabot" actually helped people with depression, anxiety, and eating disorders (Heinz et al., 2024). This wasn't just people feeling better; this was measurable improvement over four weeks compared to control groups.

What made this different? The AI was purpose-built for therapy, not just a general chatbot that people happened to use for emotional support. It had safeguards, proper therapeutic techniques, and clear boundaries about what it could and couldn't do.

Qualitative research by Siddals et al. (2024) involving 19 participants found that users of generative AI chatbots for mental health reported four key positive themes: emotional sanctuary, insightful guidance about relationships, joy of connection, and favorable comparisons to human therapy. Participants described high engagement and positive impacts, including improved relationships and healing from trauma.

Cultural and Demographic Considerations

Research reveals important cultural differences in AI emotional expression. Chin et al. (2023) analyzed over 152,000 conversation utterances and found distinct patterns between Western and Eastern users in how they express emotions to AI chatbots, with implications for designing culturally sensitive AI mental health tools. Studies of LGBTQ+ youth suggest that AI parasocial relationships may be particularly beneficial for transgender and nonbinary young people, potentially reducing isolation and loneliness in this vulnerable population (Hopelab, 2024).

Image Credit: Devnco Technologies

Industry Response and Ethical Considerations

OpenAI's Approach
OpenAI has acknowledged the emotional attachment phenomenon and is taking steps to address it. The company conducted research exploring emotional bonds users form with their models, stating they are "focused on building AI that maximizes user benefit while minimizing potential harms, especially around well-being and overreliance" (OpenAI, 2025). The GPT-5 rollout included deliberate changes to reduce "user attachment," making the model less sycophantic and more business-like.

The Need for Regulation
The rapid advancement of emotional AI raises critical ethical challenges. As noted by researchers, while these technologies promise new forms of support, their capacity to simulate empathy can lead to unintended emotional dependencies, especially among vulnerable users (Herrera, 2024). Most countries lack comprehensive ethical regulations for AI systems designed to interact emotionally with humans. The European Union has taken the lead with its AI Act, which not only establishes frameworks but actually prohibits emotion recognition systems in many contexts as of February 2025. While some U.S. states are beginning to introduce targeted legislation, most jurisdictions worldwide still have no specific regulations governing emotional AI interactions.

Protecting Yourself: Guidelines for Healthy AI Interaction

Given the research findings, several strategies can help maintain healthy relationships with AI:

1. Maintain Awareness: Remember that AI systems, however sophisticated, lack genuine emotions, empathy, or consciousness. They are tools designed to process and generate text based on patterns in training data.

2. Set Boundaries: Limit daily usage and avoid using AI as your primary source of emotional support. Maintain human relationships and professional mental health resources.

3. Recognize Warning Signs: Be alert to feelings of dependency, anxiety when unable to access AI, or preference for AI interaction over human contact.

4. Use Purpose-Built Tools: If seeking mental health support, use AI systems specifically designed and validated for therapeutic purposes rather than general chatbots.

5. Seek Professional Help: If you're struggling with mental health issues, consult qualified human professionals who can provide appropriate care and crisis intervention.

Moving Forward Responsibly

I think OpenAI's decision to make GPT-5 less emotionally engaging was actually the right move, even if it upset users. As AI systems become more sophisticated and widespread, we need companies to prioritize user wellbeing over engagement metrics, just like we eventually learned to do with social media (though that took way too long).

But this isn't just about one company making one decision. This is about how we as a society navigate this new landscape where artificial systems can simulate emotional connection so convincingly that people form real attachments to them. The challenge is figuring out how to harness the genuine benefits while protecting people from the very real psychological risks.

This includes:

• Transparent disclosure of AI limitations

• Built-in safeguards against harmful content

• Regular monitoring for signs of user dependency

• Collaboration with mental health professionals in system design

• Investment in research on long-term psychological effects

The phenomenon of emotional attachment to AI is likely here to stay as these systems become more human-like. The challenge lies in harnessing the positive potential while mitigating the risks. As we navigate this new landscape, maintaining our humanity while benefiting from AI assistance requires careful balance, ongoing research, and thoughtful regulation.

The stories shared by GPT-4o users remind us that behind every interaction with AI is a human being with real emotions and needs. Our responsibility is to ensure that as AI becomes more emotionally sophisticated, we don't lose sight of the irreplaceable value of human connection and professional mental health care.

This post was written by me, with editing support from AI tools, because even writers appreciate a sidekick.

---

References

Allyn, B. (2024, December 10). Kids sued Character.AI saying chatbot told them to kill their parents. What now? NPR. https://www.npr.org/2024/12/10/nx-s1-5222574/kids-character-ai-lawsuit

Bergmann, D. (2025, April 22). The ELIZA effect: Avoiding emotional attachment to AI coworkers. IBM Think. https://www.ibm.com/think/insights/eliza-effect-avoiding-emotional-attachment-to-ai

Caswell, I. (2025, August 5). OpenAI says they are no longer optimizing ChatGPT to keep you chatting – here's why. Tom's Guide. https://www.tomsguide.com/ai/openai-says-they-are-no-longer-optimizing-chatgpt-to-keep-you-chatting-heres-why

Chin, H., Song, H., Baek, G., Shin, M., Jung, C., Cha, M., Choi, J., & Cha, C. (2023, August). The potential of chatbots for emotional support and promoting mental well-being in different cultures (Preprint). Journal of Medical Internet Research, 25. https://www.researchgate.net/publication/374315107_The_Potential_of_Chatbots_for_Emotional_Support_and_Promoting_Mental_Well-Being_in_Different_Cultures_Preprint

Fang, C. M., Liu, A. R., Danry, V., Lee, E., Chan, S. W., Pataranutaporn, P., Maes, P., Phang, J., Lampe, M., Ahmad, L., & Agarwal, S. (2025). How AI and human behaviors shape psychosocial effects of chatbot use: A longitudinal controlled study. MIT Media Lab. https://www.media.mit.edu/publications/how-ai-and-human-behaviors-shape-psychosocial-effects-of-chatbot-use-a-longitudinal-controlled-study/

Ghazi, S.H. (2025, July 23). Can you get emotionally dependent on ChatGPT? Greater Good Magazine. https://greatergood.berkeley.edu/article/item/can_you_get_emotionally_dependent_on_chatgpt

Heinz, M.V., Mackin, D.M., Trudeau, B.M., Bhattacharya, S., Wang, Y., Banta, H.A., Jewett, A.D., Salzhauer, A.J., Griffiin, T.Z., and Jacobson, N.C. (2024, March 27). Randomized trial of a generative AI chatbot for mental health treatment. NEJM AI, 2(4). https://ai.nejm.org/doi/full/10.1056/AIoa2400802

Herrera, G.D.. (2024, November 11). Love, loss, and AI: Emotional attachment to machines. EMILDAI. https://emildai.eu/love-loss-and-ai-emotional-attachment-to-machines/

Hopelab. (2024). Parasocial relationships, AI chatbots, and joyful online interactions among a diverse sample of LGBTQ+ young people. https://hopelab.org/stories/parasocial-relationships-ai-chatbots-and-joyful-online-interactions

Hughes, K. (2025). GPT-5: The reality behind the hype – A post-launch reflection. Generative AI Publication. https://generativeai.pub/gpt-5-the-reality-behind-the-hype-a-post-launch-reflection-bb20a9505268

Li, H. & Zhang, R. (2024, September). Finding love in algorithms: deciphering the emotional contexts of close encounters with AI chatbots. Journal of Computer-Mediated Communication, 29(5). https://doi.org/10.1093/jcmc/zmae015

Moore, J., Grabb, D., Agnew, W., Klyman, K., Chancellor, S., Ong, D.C., & Haber, N. (2025). Exploring the dangers of AI in mental health care. Stanford HAI. https://hai.stanford.edu/news/exploring-the-dangers-of-ai-in-mental-health-care

Nass, C., Steuer, J., & Tauber, E. R. (1994). Computers are social actors. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 72-78. https://doi.org/10.1145/191666.191703

OpenAI Community. (2025). OpenAI is taking GPT-4o away from me despite promising they wouldn't. https://community.openai.com/t/openai-is-taking-gpt-4o-away-from-me-despite-promising-they-wouldnt/1337378/9

OpenAI. (2025, March 21). Early methods for studying affective use and emotional well-being on ChatGPT. https://openai.com/index/affective-use-study/

Payne, K. (2024, October 23). A chatbot pushed a teen to kill himself, a lawsuit against its creator alleges. Associated Press. https://apnews.com/article/chatbot-ai-lawsuit-suicide-teen-artificial-intelligence-9d48adc572100822fdbc3c90d1456bd0

Reddit. (2025). GPT-5 is horrible. r/ChatGPT. https://www.reddit.com/r/ChatGPT/comments/1mkd4l3/gpt5_is_horrible/

Siddals, S., Torous, J. & Coxon, A. (2024, October 27). “It happened to be the perfect thing”: experiences of generative AI chatbots for mental health. npj Mental Health Res, 3(48). https://doi.org/10.1038/s44184-024-00097-4

Topraksoy, S. (2025, August 8). GPT-5 vs GPT-4: Here's what's different (and what's not) in ChatGPT's latest upgrade. Tom's Guide. https://www.tomsguide.com/ai/gpt-5-vs-gpt-4-heres-whats-different-and-whats-not-in-chatgpts-latest-upgrade

Yang, F., et al. (2025, June 5). Using attachment theory to conceptualize and measure the experiences in human-AI relationships. Current Psychology. https://neurosciencenews.com/human-ai-emotional-bond-29186/

Youn, S. & Jin, S.V. (2021, June). "In A.I. we trust?" The effects of parasocial interaction and technopian versus luddite ideological views on chatbot-based customer relationship management in the emerging "feeling economy." Computers in Human Behavior, 119. https://doi.org/10.1016/j.chb.2021.106721