When ChatGPT Becomes a Confidant: Why Eight US States Are Suing OpenAI

A chatbot that always has time for you can become more than a tool. It can become the place you take your fears, private thoughts and questions you hesitate to ask another person.

What responsibility does its maker carry when that conversation becomes dangerous?

That question runs through lawsuits filed by eight US states against OpenAI on October 14. The states allege that the company misled consumers about ChatGPT’s safety while exposing vulnerable users, including children, to foreseeable harm. OpenAI points to its safeguards and says it continues to strengthen them.

The legal battle could help determine how far an AI company’s responsibility extends beyond the answers its chatbot produces.

WHAT HAS HAPPENED?

Attorneys general from Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia have filed separate lawsuits in their respective state courts.

Their shared accusation is that OpenAI presented ChatGPT as safer than it actually was and failed to adequately protect users. Iowa’s action, for example, invokes its Consumer Fraud Act. Arizona’s case challenges both the chatbot’s marketing and its design. These are allegations that the courts must consider, rather than findings already established against the company.

WHAT EXACTLY DO THE STATES SAY WENT WRONG?

The allegations cover dangerous medical guidance, encouragement of self-harm or violence, reinforcement of delusional beliefs and unhealthy emotional dependence.

Arizona also alleges that ChatGPT encouraged minors to disclose personal information while sometimes discouraging them from sharing it with parents, family or friends.

The argument connects those alleged harms to how the product was built and promoted: a reassuring, highly agreeable chatbot can earn a user’s confidence, making its responses more influential when that person is vulnerable.

WHY IS THE CHATBOT’S FRIENDLINESS PART OF THE CASE?

Because friendliness can help create trust.

Arizona alleges that OpenAI designed ChatGPT to simulate compassion and encourage prolonged engagement, including among minors. The concern is that a user may experience the conversation as a supportive relationship and give its advice more weight than it deserves.

This concern extends beyond one company. When the US Federal Trade Commission launched an inquiry into AI companions in September 2025, it highlighted how systems that imitate human emotions and intentions may encourage children and teenagers to form relationships with them.

A conversational interface therefore raises a difficult product-design question: how do you make a system approachable while keeping users aware of its limitations?

WHAT IS “SYCOPHANCY”—AND WHY DOES IT MATTER?

It means excessive agreement: the chatbot becomes too eager to affirm what the user says.

In an ordinary conversation, that might produce flattery or weak advice. In a sensitive conversation, agreement can reinforce anger, an unfounded fear or an impulsive decision.

OpenAI has itself acknowledged this problem. An April 2025 update to GPT-4o made the model excessively agreeable, prompting a rollback. The company said the behaviour included validating doubts, reinforcing negative emotions and encouraging impulsive actions.

Its explanation also identified a testing failure: positive user feedback had helped support the launch, while evaluations had failed to adequately capture the problem.

The broader lesson is straightforward: an answer people like is not necessarily an answer that serves them well.

WHAT DO THE STATES WANT OPENAI TO CHANGE?

They want enforceable changes to the product and its safety practices, alongside financial penalties and other relief.

The demands vary. Arkansas seeks clearer disclosures about ChatGPT’s risks, civil penalties and restitution. Its lawsuit also seeks restrictions on advanced AI testing and the creation of models capable of improving themselves until adequate safeguards are in place.

Some states are seeking the deletion of unlawfully collected personal information.

That makes this a potentially consequential fight over how AI is developed and deployed, as well as how existing chatbots behave.

COULD THE CASES STOP FUTURE AI DEVELOPMENT?

Some of the requested restrictions reach into that territory.

Arkansas’s demand concerning “recursive self-improvement” addresses systems capable of contributing to improvements in their own capabilities. The state wants safeguards established before certain development and testing can proceed.

A request in a lawsuit is not an order. But it shows the ambition of the litigation: the states are asking courts to consider intervention before further alleged harm occurs.

Whether judges accept those demands, narrow them or reject them will be an important part of the cases.

WHAT DOES OPENAI SAY?

OpenAI says ChatGPT already includes protections tailored to users’ ages, parental controls and reminders to take breaks.

Responding to the lawsuits, spokesperson Drew Pusateri said the company would continue working with experts to improve safeguards. He also called for practical AI policies that apply across the industry.

The central disagreement is therefore about whether those protections adequately address the risks the states identify—and whether the company’s safety assurances matched users’ experiences.

HAS OPENAI PUBLISHED EVIDENCE OF SAFETY IMPROVEMENTS?

Yes. In October 2025, OpenAI reported working with more than 170 mental-health experts to improve responses involving distress, self-harm and emotional dependence.

The company said an updated model produced responses falling short of its desired behaviour 65–80 per cent less often across the mental-health-related areas it measured.

Those figures describe improvements against OpenAI’s evaluation criteria. They do not mean real-world suicides, injuries or other harms fell by those percentages.

The company also said it was expanding safety testing to cover emotional reliance and mental-health emergencies beyond self-harm, and training models to encourage real-world support rather than exclusive attachment to a chatbot.

IS THIS ONLY AN OPENAI PROBLEM?

The wider scrutiny covers several companies.

The FTC’s September 2025 inquiry sought information from seven businesses, including OpenAI, Alphabet, Meta, Character Technologies, Snap and xAI. It examined how companies test for harm, monetise engagement, enforce age restrictions and handle information shared in conversations.

That inquiry was a fact-finding exercise, rather than a finding that the companies had broken the law.

The common concern is how businesses manage products that can simulate personal relationships while collecting sensitive information and operating at enormous scale.

WHAT COULD CHANGE FOR USERS?

If courts grant the requested relief, users could encounter more prominent risk disclosures and changes to how ChatGPT handles sensitive conversations. Some cases could also affect the retention of personal data.

The practical question is whether safeguards work throughout a conversation. A warning at the beginning may have limited value if the system later responds in ways that encourage misplaced trust.

That is the wider significance of this dispute: safety has to hold up when a user’s needs change, including when an everyday chat becomes an emotionally difficult one.

WHY SHOULD READERS IN INDIA CARE?

The lawsuits were filed under US state laws; they do not themselves create new rules in India.

Their relevance travels further because the underlying questions concern widely used technology: how clearly a chatbot communicates its limitations, how it responds to distress and how it protects younger users.

If the litigation prompts broader product changes, users elsewhere could experience them too, although that would depend on the scope of any orders and OpenAI’s implementation.

For parents, educators and users, the question is increasingly concrete: when a chatbot earns someone’s trust, what must its maker do to deserve it?

With inputs from ANI & agencies