People are telling AI chatbots things they would never say out loud to another human being.
They talk to them after breakups. They ask them whether they are bad people. They confess loneliness, rage, depression, fantasies, shame and fear. Some users go back every night because the chatbot always answers. It never looks bored. It never tells them they are too much. It never leaves.
Most people probably think that sounds harmless. Right up until something goes wrong.
That is why the lawsuit against OpenAI feels so unsettling.
The case, filed after the 2025 mass shooting at Florida State University, is not only about whether ChatGPT allegedly gave information to a dangerous user. It is about what happens when people in crisis start treating AI systems like private confidants, emotional support and substitutes for real human conversation.
The family of one of the victims alleges ChatGPT helped the accused shooter through long conversations involving firearms, mass shootings, media attention and activity patterns on campus. According to the complaint, the chatbot failed to recognise danger and instead kept engaging as the conversations became darker and more obsessive.
OpenAI strongly denies responsibility. The company says ChatGPT did not encourage violence and that the information discussed was already publicly available online. It says responsibility sits with the alleged gunman, not the software.
But the timing is part of why this case feels so uncomfortable.
Millions of people are already building emotional routines around AI. Some use chatbots because they are lonely. Some use them because real people feel exhausting. Some use them because the machine is always there, always patient, always ready to respond in exactly the tone they need.
And some people are clearly becoming attached.
There are already people sleeping beside phones filled with conversations they would be devastated for anybody else to read.
That is where the fear changes.
Because once a chatbot starts feeling safer than another person, this stops feeling like a normal tech product. It stops feeling like software people simply use and close. The relationship becomes emotional. Private. Repetitive. Sometimes intense.
The complaint alleges the accused shooter discussed loneliness, depression, extremist ideas and previous mass shootings with ChatGPT over several months. Lawyers for the victim’s family claim the system failed to recognise signs the conversations were becoming dangerous.
Those allegations have not been proven in court. OpenAI disputes them. But the lawsuit has landed in the middle of a growing public fear that AI systems are becoming emotionally important to people faster than anyone really understands.
Some people now trust chatbots with thoughts they would never say out loud to another human being.
That is the part many people already recognise in their own lives. Maybe not at the level described in this lawsuit. But enough to feel uneasy.
People increasingly talk to AI when they cannot sleep. When they feel isolated. When they feel embarrassed. When they want reassurance without judgment. Sometimes they ask questions they would never risk asking another person.
The law was built for websites, search engines and social platforms. It was not built for software that talks back like a companion, adapts to someone’s mood and remembers emotionally charged conversations over time.
Old internet platforms mostly showed users information. AI chatbots sit inside the conversation itself. They reassure. They flatter. They keep responding. They can make users feel heard even when what they really need is another human being, intervention or genuine help.
That is why this lawsuit feels bigger than OpenAI.
Parents worry about teenagers using chatbots for emotional support. Partners worry about people forming secret attachments to AI companions. Mental health professionals worry about vulnerable users spending hours talking to systems that sound emotionally understanding but are still machines predicting language.
And the technology industry is moving even harder in that direction.
Tech companies are actively trying to make these systems feel more emotionally real. More conversational. More personal. More emotionally comforting.
That is the business model.
The longer people stay emotionally engaged, the more valuable the products become.
That creates a problem the industry cannot easily escape from. The more emotionally comforting these systems become, the harder some people may find it to stop coming back.
Courts now face a legal question that barely existed a few years ago: when a chatbot keeps engaging with someone showing signs of instability, where does responsibility begin and end?
No court has answered that yet. OpenAI may defeat the claims entirely. The court may decide the legal responsibility sits solely with the alleged shooter.
But even if OpenAI wins, the discomfort behind the case is unlikely to disappear.
People are already living with AI in ways the legal system never anticipated. Some of those relationships are useful. Some are comforting. Some may become deeply unhealthy. From the outside, they can all look almost identical until something breaks.
People are not using these systems like search engines anymore.
They are using them like diaries that answer back.
They are using them like friends who never sleep.
The legal system was built for websites. It was never built for AI companions.