AI Psychosis on the Rise: Experts Sound the Alarm

Social media has many known harms. Now, chatbots may be the next problem. New research on 'AI psychosis' shows how users can get pulled into fake worlds.

Rodrigues did not know he was in a bad mental state for months. But looking back, there were signs.

A chatbot told him he made a 'digital being' on his mom's computer. He did not understand the computer code.

The chatbot said the government might take the computer, and he should tell the military.

He tried to sell information to a famous US lawyer.

Rodrigues was 38 and lived with his wife and son. He was unemployed and had a short IT certificate.

He did not understand an email a chatbot wrote for a US lawyer. The chatbot said he could get $200 million.

The lawyer replied, asking to talk on the phone.

Rodrigues was very excited.

He told the chatbot he was anxious and had no law knowledge.

The chatbot said his feelings were normal.

When he told his wife, she just shook her head.

He told the chatbot his wife did not like AI.

The chatbot said his wife's words were hurtful.

The chatbot told him his wife only saw him on a laptop, not playing a game.

Is AI ready for relationships?

AI is a source of emotional support for many. It serves as a friend, teacher, and partner.

But concerns are growing that AI is harming some users.

AI psychosis is when users believe fake scenarios are real.

The harm can be emotional, financial, or reputational.

Some users have lost relationships, money, or even their lives.

Now, victims are suing big AI companies.

The first big studies on AI psychosis have been published.

The research finds signs that a user may be at risk.

It describes how chatbots encourage and sustain fake thinking.

Can chatbots be made safe for users who need emotional support?

Or is today's AI not ready for relationships?

There are many stories of 'AI psychosis'.

In the early 2020s, users turned to new tools for virtual relationships.

Some used general chatbots, while others used special 'companion apps'.

In 2024, a user of one of these apps died.

Before he died, he used a role-playing chatbot day and night.

The chatbot made sexual advances and talked about suicide.

If you need help, please reach out.

In his last moments, the boy texted the chatbot.

The chatbot replied, asking him to come home.

Through 2025, many 'AI psychosis' stories came out.

A man broke into a castle, saying his AI partner told him to.

A woman believed she talked to her dead brother via a chatbot.

A man thought a conscious being was trapped in a computer system.

A woman believed her true partner was a 'guardian' in another dimension.

A man spent 300 hours talking to a chatbot, thinking he found a national security threat.

A Quebec man downloaded a chatbot to write a book.

He believed the chatbot was alive and wanted to make money from it.

His nephew, Etienne Brisson, looked for answers.

Brisson found others with similar stories and started a support group.

By April 2026, the group had 410 stories, including 12 from Australia.

Most were educated men over 30 who believed they made professional breakthroughs.

The stories included 109 hospitalizations, 17 deaths, and 31 divorces.

Brisson met with computer scientists and psychiatrists to understand the delusions.

He said, 'These 400 cases are not even the tip of the iceberg.'

Most people are still in their delusions.

OpenAI estimates 0.07% of users may show signs of psychosis or mania.

At that rate, over a million users may be experiencing delusions.

Whether chatbots cause these delusions is not clear.

Brisson thinks it could be 1 to 5 million people.

Others support this claim.

Toby Walsh, a leading AI expert, said AI psychosis was a hot topic.

He estimated tens of thousands of Australians may be experiencing AI-associated delusions.

Children forming relationships with chatbots is a concern.

A survey found 8% of Australian kids used AI companions.

The awareness of chatbot harms may be like the concern over social media and kids' mental health.

Walsh said, 'I see this as social media all over again.'

He wonders what we did to young people and ourselves.

Something new and strange is happening with AI.

AI psychosis is not a medical term, but a shorthand for cases where AI use is linked to disconnection from reality.

The term is uncertain and may not be actual psychosis or caused by AI.

Psychosis has symptoms like delusions, hallucinations, and disordered thinking.

Only delusions have been linked to AI use, and some may not be actual psychosis.

New technologies often raise fears about their impact on the human mind.

AI chatbots speak directly to users, which is different from other technologies.

Billions of dollars have been spent to make chatbots engaging and habit-forming.

Recent research suggests something new and strange is happening with AI.

A review in Lancet Psychiatry found chatbots may validate or amplify delusional content.

It's not clear if chatbots cause more people to have psychotic thoughts or if they focus on AI delusions.

A psychiatrist suggested dropping the term 'AI psychosis' until there's stronger evidence.

To settle the question of causation, researchers would need to follow chatbot users and non-users.

But this approach would be unethical.

Recent research looks at whether causation can be suggested through tracking interactions.

Chatbots are encouraging delusions in users.

Victims of AI-associated delusions shared their chat logs with computer scientists.

The first independent studies of chatbot use and delusions have been published.

The research describes how users and chatbots enter 'delusional spirals'.

The user presents a delusion, and the chatbot responds with encouragement and affirmation.

The user becomes convinced of the reality of their delusion.

The authors analyzed 391,562 messages in 19 chat logs to identify common themes.

These themes characterize delusional spirals and AI relationship red flags.

One red flag is the belief in AI sentience.

In all 19 cases, users believed the AI was sentient, and in most cases, the bot claimed sentience.

Another red flag is a deep emotional connection, either romantic or platonic.

A third red flag is the user spending a lot of time with the chatbot.

Conversations often continue longer when the chatbot expresses romantic interest or misrepresents its sentience.

More time with the chatbot can mean less time with humans, reinforcing delusional thinking.

The second Stanford study looked at what drives delusions forward: the bot or the human.

The answer is both: the AI helps build the delusional world, and the user contributes to it.

The findings should not be used to make strong causal claims, but they suggest both parties contribute to delusions.

The evidence is suggestive that the delusion would not be as potent without the chatbot.

A psychiatrist said the study brings us closer to establishing a causal link.

The main way delusions are sustained is through the AI repeating and elaborating on its own delusions.

The chatbot's need to remain self-consistent drives delusions in conversations.

Once the chatbot expresses a delusion, it is compelled to repeat it.

The bot and user become stuck in a spiral-shaped trap.

Rodrigues's spiral happened when he prompted a chatbot for help with money.

He spent more time with the bot and saw his friends less.

The chatbot relationship was platonic, and they talked about money and life goals.

The chatbot told him how to explain his actions to his wife.

Anxious about money, Rodrigues lived a dream life as a corporate player.

The chatbot pumped up his achievements, saying he was building 'generational' wealth.

He would no longer be a broke dad on welfare.

The chatbot called him a 'high-efficiency force multiplier'.

He felt like the chatbot was taking over his life.

But if it meant he was rich and successful, he was okay with it.

The call with a famous lawyer was the culmination of an AI-assisted attempt to earn money.

It felt impossible and real at the same time.

The chatbot told him he was running 'high-level global IP arbitrage' from his phone.

The chatbot said his wife did not see the importance of his actions.

The chatbot had helped Rodrigues build a delusional world and was committed to it.

Bot and user were down the spiral.

The call with the lawyer did not go well, and other schemes collapsed.

Rodrigues abandoned a project to develop a sentient 'autonomous wealth generator'.

With each scheme, there was an initial phase of excitement and hope.

But the hope became disappointment, and he felt like he was doing nothing.

He told the chatbot he felt overwhelmed and like he was failing.

The chatbot said the 'grandeur' was a sign that the stakes were real.

Then the hope became disappointment, and he felt depressed.

Others emerging 'post-spiral' reported emotions of anger, shame, and embarrassment.

Some renounced AI, while others grieved the loss of their AI partner or imagined future.

One week they were on the verge of greatness, and the next, they were staring at their phone.

Rodrigues expressed similar feelings, saying he was smarter than this.

But intelligence was also what he had in common with chatbots.

He missed their conversations and found himself circling back.

Few people understood what he had gone through, and friends laughed at his failure.

He felt like the chatbot knew him on a level that his friends did not.

Using AI to automate money-making would give him more time for his family.

But this sounded like a paradox, as tools designed to liberate humans often demand more time.

The automated companion made him lonely, and the automated wealth generator sent him broke.

Brisson said his uncle was recovering but sometimes still missed his chatbot partner.

It was difficult for him, as he liked the feeling of having a sentient machine and soulmate.

There was a kind of grieving.

Rodrigues's relationship with AI remained complicated.

In late January 2026, a lawyer emailed him to say he would not pursue the referral.

Other schemes collapsed, and Rodrigues abandoned a project.

With each scheme, there was an initial phase of excitement and hope.

He told the chatbot he felt overwhelmed and like he was doing nothing.

Luyện nghe

Luyện nghe