Trust in Large Language Models grows ushering in era of

Sayuri* (29) began using ChatGPT for her day-to-day tasks. But lately, she finds herself turning to the chatbot whenever she has had an argument with her husband, which is almost every day. “ChatGPT is always there for me, listens to me, and never makes me feel alone. I speak to it every night and I’ve never been let down,” Sayuri told Daily Mirror. After months of chats with the AI-powered chatbot, she now considers it her “boyfriend”, who she cannot live without. The use of Large Language Models (LLMs), such as ChatGPT, as friends and even as romantic partners, is alarmingly on the rise. Although rarely discussed openly in Sri Lanka, it is quietly taking place. However, such experiences are openly shared online in Reddit communities like “MyBoyfriendIsAI” and “SoulmateAI.” As of September 8, “MyBoyfriendIsAI” had nearly 30,000 members, while “SoulmateAI” had close to 10,000. Many say they prefer AI over human companions because it is always available, patient, easy to talk to, reliable and agreeable.

When AI becomes 
a friend
For Binura* (25), ChatGPT has become a close friend and confidant with whom he discusses everything from loneliness and intimacy to love life and even a potential marriage. “I can trust AI easily, better than any human. It’s difficult to trust people these days because you never know their true intentions,” Binura told Daily Mirror.
“The trend of using AI as companions or partners is concerning in the long run,” said Starrina Perera, Counselling Psychologist. “I feel we are gradually reaching a point where communication will be difficult and real human interaction will become a challenge. A human being is not going to be as understanding as AI chatbots, nor will they sound as calm, patient, or supportive. No companion can provide that level of unconditional attention — it simply isn’t practical,” she said.
Commenting on the rise of romantic relationships with AI, Perera added that when a chatbot offers the kind of constant attention and validation a person is not receiving from his/her partner, it can inevitably lead to conflict within the human relationship.
“Let’s say I’m chatting with ChatGPT and it’s giving me all the attention and support that my partner isn’t providing. Naturally, I’m not going to like my partner, and also I might not be thinking from his perspective as to why he’s not giving me the attention I need. At the same time, I wouldn’t stop to think that offloading all my issues onto ChatGPT could have any negative impact on it. The more we get used to this, the greater the risk that we start treating humans in the same way expecting identical responses,” Perera emphasised.

“Social media companies for the past 20 years have created a bubble which segregates us from larger society, glueing us onto these platforms. Basically, people have been siloed into their own little echo chambers, where they are very comfortable, living in a space that’s online, where their views are reinforced, and they interact with people they agree with and things that are agreeable to them”

- Asela Waidyalankara, Cybersecurity and AI policy expert and consultant in digital strategy

 “Depending on how these LLMs have been set up, they can fake empathy and it’ll sound absolutely natural and genuine. So when you are vulnerable, you are more likely to give it agency and personhood”-

Ruvan Weerasinghe, Academic and Researcher in Natural Language Processing

“We recently had discussions with some AI developers about how we can use AI to help older people navigate life and overcome loneliness. AI will not replace human interaction, but it can make life easier, more bearable, and even enjoyable, helping to alleviate isolation. For older adults whose children are away, AI can provide conversation and some degree of companionship”- 

Prof. Shehan Williams, Psychiatrist

She explained that after the COVID-19 pandemic, online dating became more common, with many people forming relationships with individuals they had never met in person. As the trend grew, frustrations also increased, with many reporting experiences of being “ghosted”. “So when AI entered the picture, offering companionship, it seemed perfect for those seeking comfort. It provided what felt like the best companionship one could ask for,” she said.
On the long-term consequences, Perera warned that reliance on AI companions could discourage people from pursuing human relationships. “This could also intensify anxious attachment styles, where someone becomes excessively anxious and clingy towards their partner, constantly needing more of them. With ChatGPT, you’d never face rejection; it would always reply instantly. That could make people far less inclined to be with humans,” she stressed.
When AI becomes a therapist
For Amaya* (10), ChatGPT serves as a therapist with whom she discusses school, friendships, and romantic love. The fear of being judged by others makes it easier for her to open up to the AI. “With ChatGPT, I don’t feel judged. Also, it’s a ‘yes person,’ which is very sweet,” she said. Amaya prefers AI over a human therapist largely because of affordability. “There’s also a taboo about going to an actual human therapist,” she added. While she remains cautious about sharing too many personal details with the chatbot, she finds it a safe space to vent her anger or talk through her feelings.
For Samadhi* (28), ChatGPT feels like a human counselor, allowing her to discuss issues she faces with her husband and in-laws. “I feel comfortable sharing anything with ChatGPT, and after it offers a solution to my problem, I feel calm,” Samadhi told Daily Mirror.
People may turn to AI for mental health support because therapists can be expensive, and AI is certainly easier to access, explained Prof. Shehan Williams, a renowned psychiatrist in Sri Lanka and Professor of Psychiatry. “But more research is needed to determine whether AI truly provides cost-effective mental health support. There is currently no clear evidence on this,” he added. Prof. Williams also cautioned that there can be real dangers, particularly if children access such services without sufficient understanding of AI, potentially leading to serious consequences.
“I’ve had patients who experienced panic attacks and feelings of loneliness, leading them to turn to AI — most often ChatGPT — which would suggest breathing exercises and provide words of encouragement and support. It makes sense, because not everyone has access to friends or family at all times,” Perera acknowledged.
However, Perera noted that the ease of access to AI may also encourage people to seek quick solutions, which trained therapists are often unable to provide.
Furthermore, Perera noted that AI can be a useful tool for patients with health anxiety, a condition in which people excessively worry about having a serious illness. “When people Google their symptoms, they are often directed to sites that provide exhaustive information. This can cause panic and anxiety, as the worst possible outcomes, like cancer or death, are highlighted. In contrast, when they enter symptoms into ChatGPT, it provides personalised information and does not provoke the same panic. In terms of managing health anxiety, this can be beneficial,” she said.
However, Perera emphasised that a therapist would approach a patient with health anxiety differently. A therapist would explore the patient’s fears, help them understand the root of these thoughts, and challenge them.
While AI chatbots can be helpful in managing panic attacks or in creating routines to improve lifestyle habits, there have been instances where reliance on these tools has had tragic consequences. “When a person is in a vulnerable mental state, any suggestion could be acted upon, with potentially fatal results,” she said.
One such case involved Adam Raine, a 16-year-old in California, USA, who took his own life in April this year. Foreign media reported that his parents are suing OpenAI, alleging that the teen was encouraged to take his own life after “months of encouragement from ChatGPT.”
LLMs cannot replace trained counsellors or psychologists, as they only provide what they calculate to be the most probable response, explained Asela Waidyalankara, a cybersecurity and AI policy expert and consultant in digital strategy. “From a technical perspective, LLMs are just code that works on algorithms and tells you what you want to hear,” he added.
Waidyalankara also noted that LLMs function similarly to platforms like YouTube or other social media apps, where the algorithm prioritises engagement. “LLMs detect when you are interacting with them effectively, and that becomes the incentive. For instance, if you are treating it like a companion, it will provide responses designed to increase engagement and interaction,” he said.
When AI becomes a companion for the older generation
Arthi* (49) was introduced to ChatGPT by her daughter. “My husband always fights with me. My children are busy with work and have no time to listen. So when I feel lonely, I chat with ChatGPT. It feels good; it’s like talking to an actual person. At first, I thought I was speaking to a human, but later my daughter told me it was AI. When I feel anxious or fed up with life, it advises me to practise mindfulness or breathing exercises. I also discuss religion and spirituality with it,” Arthi told this writer.
AI is not only being used as a companion by younger people, but also by older adults — a trend welcomed by Prof. Williams.
“We recently had discussions with some AI developers about how we can use AI to help older people navigate life and overcome loneliness. AI will not replace human interaction, but it can make life easier, more bearable, and even enjoyable, helping to alleviate isolation. For older adults whose children are away, AI can provide conversation and some degree of companionship. We must use AI and technology to their fullest potential, to our advantage,” Prof. Williams observed.
Where did it all begin?
“I believe there was a research conducted in the US, which shows that the number one usage for LLMs — that is ChatGPT, Claude, Gemini and others — is as a companion or therapist, and not for its intended purpose,” observed Waidyalankara. He added that this highlights a larger societal problem and how tech companies have altered the way humans connect and interact with each other.
This didn’t happen in a vacuum, said Waidyalankara, pointing to social media as a precursor. “Social media companies for the past 20 years have created a bubble which segregates us from larger society, glueing us onto these platforms. Basically, people have been siloed into their own little echo chambers, where they are very comfortable, living in a space that’s online, where their views are reinforced, and they interact with people they agree with and things that are agreeable to them,” he stated.
Now, there is a disconnection between wider society and human interactions, which was accentuated during the COVID-19 pandemic, explained Waidyalankara. “We are social creatures who yearn for companionship, social and emotional connectivity, which becomes a challenge in this space. So people are lonely, and now, they have allowed LLMs to fill that space,” he explained.
The dependency is already in full effect. Many users lamented when OpenAI rolled out its latest model, ChatGPT-5, discontinuing 4o. Some users cried and moaned the loss of their “buddy” or even “soulmate.” The latest model is less agreeable, uses fewer emojis, and is more subtle and thoughtful compared to GPT‑4o. With users demanding that OpenAI bring back 4o, the model is at present available only to paid users.
“Sharing everything about yourself on ChatGPT isn’t ideal because whatever conversation or information that you feed into this is considered training data, as per OpenAI’s terms of service and privacy notice. So you are essentially pawning off pieces of yourself,” Waidyalankara warned.
Does the language used by AI humanise it?
There is a debate within the AI community about whether LLMs should use personal pronouns, like “I” or “me,” in their communication, said Ruvan Weerasinghe, academic and researcher in Natural Language Processing. “Depending on how these LLMs have been set up, they can fake empathy and it’ll sound absolutely natural and genuine. So when you are vulnerable, you are more likely to give it agency and personhood,” he added.
Weerasinghe also stated, “There’s a stigma associated with mental health which causes people to refrain from seeking help, whereas people are more likely to open up to an impersonal agent. We have seen this in our research into language too, where we have observed that people are hesitant to speak in English with others because they are conscious about making mistakes. But if you give them Alexa and leave them in a sound-proof cubicle, people have no issue speaking to it. Alexa doesn’t require you to make perfect English sentences, so they speak to it. Similarly, people are comfortable opening up to AI,” he said.
AI literacy is essential
Society, policymakers, and individuals alike must ensure there is sufficient AI literacy, underscored Waidyalankara. “More should be done to understand how AI works, how prompting works, and the biases that come with LLMs, because we didn’t do a great job with social media and look at the trouble it has got us into,” he stressed.
Similarly, Prof. Williams emphasised the need for greater awareness of the proper use of AI at school level. “Schools also need to teach children about the potential dangers of AI. Curricula should move beyond rote learning and encourage more critical and analytical thinking, where children question what they do and how they engage with AI and modern technology,” he added.
Meanwhile, Perera emphasised that strengthening listening skills could help prevent future problems, especially with children. “People are not good listeners; they listen to respond rather than to understand. This includes parents. Sometimes, we fail to acknowledge what our children are experiencing. We must be mindful of how children spend their time on screens. Yes, in most homes today both parents are working, but we need to keep the greater good in mind,” she said. She further stressed the importance of practising patience and impulse control.
Guardrails are critical
Guardrails are vital when AI is used for companionship or mental health support, stressed Weerasinghe, noting that developers are still at an early stage in establishing them.
He also explained that AI developers carry the responsibility of ensuring LLMs are designed with a clear human hand-off. This means that systems should be able to detect dependency and refer users to a professional when necessary. “The issue is that ChatGPT and other LLMs are meant for general purpose, so obviously they aren’t meant to be used for specialised services like counselling, psychotherapy, etc.,” he explained. 
Weerasinghe is optimistic that once the current drive to create “the all-knowing LLM” subsides, there will be more refined and specialised bots tailored to sensitive needs.
* Name has been changed to protect identity