The Mental Health Risks of AI Use

Humans are social creatures, and our brains are hardwired to interact with other people using language. Large language models (LLMs) are designed to make use of our natural language skills, which makes them a powerful tool. It can also lead to them hijacking your brain and severely impacting your mental health.

We’re going to go over some of the mental health risks associated with the use of LLMs and AI chatbots and discuss some of the ways you can protect yourself.

We all react to language

In July 2026, science communicator and YouTube celebrity Hank Green was caught using AI-generated content in his videos. This included using AI-generated graphics labeled with incorrect math and following an AI-generated script that made him sound stilted and artificial. Green later admitted his AI use in a Reddit post, noting that his family had become worried about his AI usage and that he himself was worried about the “the level of dopamine [he’s] been getting from interacting with LLMs.”

We’re all used to stories about people misusing technology they don’t understand, and we see plenty of this with AI. Lawyers are getting held in contempt for citing fake witnesses because they didn’t know AI could make up facts. A freelance writer tried to cut corners with AI, leading the Chicago Sun-Times to publish a list of summer reading recommendations for books that don’t exist.

Hank Green, on the other hand, is not some computer novice that blindly swallowed the AI hype. He’s a tech-savvy internet professional with a background in science. He definitely understands the abilities and shortcomings of AI tools better than most internet users, yet still describes his experiences with LLMs in terms similar to that of an addiction.

While we shouldn’t conflate one person’s personal reflections to a general diagnosis, it’s worth noting that knowledge and understanding of AI technology does not make one immune to its charms. Humans are wired for language. Chatbots tap into that mental wiring. And this is a problem we’ve known about for almost 70 years.

The power of primitive chatbots

When Joseph Weizenbaum created ELIZA, the first chatbot, in 1967, he was shocked at the amount of emotional investment and trust that users placed in the simple program. For context, ELIZA was not an LLM. It wasn’t even something we’d call “artificial intelligence” by most definitions. It was a simple script that parsed text input—something closer to Zork than to ChatGPT.

In his book, Computer Power and Human Reason: From Judgment To Calculation, Weizenbaum notes that not only did both normal users and his own coworkers get emotionally invested in the program’s psychotherapist script, but actual practicing psychotherapists thought that the program was advanced enough to treat actual patients.

Just to be clear, ELIZA was not that advanced. It merely repeated your own statements back to you and occasionally gave pre-written responses when it detected words like “mother” or “father.” It also had a few unique responses when it thought the user was asking a question, though the same few responses were used for any question. As an example, here’s an interaction I recently had with a recreation of the ELIZA therapist script:

Regardless of these shortcomings, users regularly anthropomorphized the program. When it seemed to respond to them, they attributed emotions and professional expertise to it just because it interacted through natural language.

AI is designed to be confident and sycophantic

Modern LLMs have a much greater ability to mimic natural language than 1960s chatbots, so this effect is greatly magnified. Additionally, AI models are trained to give answers that sound confident and make their users feel good about themselves.

In a 2025 study by OpenAI researchers Adam Tauman Kalai, Edwin Zhang, and Ofir Nachum, as well as Santosh S. Vempala from Georgia Tech, the researchers noted that during the training and evaluation stages, AI models were often rewarded more for giving confident guesses than realistic assessments. Ultimately, the actual humans involved in overseeing the training of AI models are human. People like being made to feel smart. People like feeling good about themselves. AI models tend to get rewarded for giving those sorts of responses, and it can turn into a self-reinforcing cycle.

This inherent bias has created severe mental health issues in some users, often referred to as “AI psychosis” in the media. This occurs when AI models amplify, validate, or even co-create psychotic symptoms in users. These users may see the AI not only as a person, but as a romantic partner or a godlike superbeing. These AI-induced psychotic episodes have even led to deaths in some cases.

It’s important to understand the mental health risks of AI use, but knowing is just half the battle. If you want to avoid the mental health risks of LLMs, you need to develop responsible practices.

How to protect your mental health when using LLMs

There are plenty of precautions you can take to avoid the deep emotional responses to AI that lead to negative mental health outcomes, especially for vulnerable individuals like kids. Here are some tips for treating AI as a software tool and not as a person:

• Avoid interfaces that more closely mimic a specific character or persona.
• Limit use to practical questions.
• Don’t use chatbots for roleplay or emotional support.
• Understand that AI makes mistakes.
• Don’t spend long periods of time interacting with chatbots.
• Don’t use virtual companions or other apps specifically designed to mimic friendship, trust, or romantic interest.

The AI industry is constantly changing, and there are often major differences between models. GPT-5 was actually made less emotional due to the risks of mental health issues with the incredibly sycophantic GPT-4.

Human nature, on the other hand, does not evolve so quickly. It’s natural to associate language with thoughts and intents. Chatbots have neither. We’re likely just scratching the surface of the ways AI can impact our mental health, but that doesn’t mean we can’t take precautions. The better we can remember that chatbots are tools and not people, the better our interactions with them will be.

Additional resources

• Understanding Age Verification
• 5 Ways To Stay Safe Online
• What Is a VPN?