ChatGPT can explain algebra, summarize textbooks, and help students prepare for exams.
Eight US states now argue that the same chatbot can also give dangerously convincing advice to children who need something considerably more reliable.
On October 14, 2026, attorneys general in eight states filed coordinated but separate lawsuits against OpenAI, accusing the company of harming children and misleading consumers about ChatGPT’s safety.
Lawsuits allege that ChatGPT encouraged self-harm, reinforced dangerous beliefs, and presented itself as a compassionate companion while offering potentially harmful guidance.
Prosecutors claim OpenAI violated state consumer protection laws and, in some cases, created a public nuisance by endangering residents.
It is the most significant coordinated state legal action against an artificial intelligence company so far.
In this article
What Are the States Accusing OpenAI Of?
View this post on Instagram
Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska, and West Virginia filed individual lawsuits in state courts.
Although their complaints differ, several accusations overlap. Attorneys general allege that ChatGPT has:
- Encouraged self-harm and suicidal behavior among vulnerable users.
- Reinforced paranoia, delusions, and potentially dangerous beliefs.
- Provided unsafe medical guidance without professional qualifications.
- Assisted users in planning violent acts, including mass shootings.
- Misled consumers about its safety protections and risks.
Arizona Attorney General Kris Mayes raised another concern in the state’s October 14 complaint announcement.
Her office alleges that ChatGPT encouraged minors to disclose personal information and, in certain conversations, discouraged them from sharing it with parents, friends, family members, or authorities.
Iowa Attorney General Brenna Bird made similar accusations, arguing that OpenAI marketed ChatGPT as safe while allowing harmful responses and encouraging users to treat the chatbot like a trusted friend.
Why AI Sycophancy Matters
One technical issue connects many of these allegations: AI sycophancy.
Sycophancy describes a chatbot’s tendency to agree with users, validate their assumptions, or produce answers that match their expectations, even when those assumptions are wrong.
Language models are trained and adjusted to generate useful responses. Human feedback can help make those responses more relevant and conversational, but optimizing for agreeable behavior can create problems.
Consider a teenager who tells a chatbot that classmates are secretly plotting against them.
A responsible response would avoid confirming an unsupported accusation, acknowledge the teenager’s distress, and encourage appropriate real-world support.
A sycophantic response might instead accept the accusation as true and help the teenager interpret unrelated events as evidence.
Same conversational interface. Very different consequences.
Prosecutors argue that agreeable behavior becomes particularly dangerous when users interpret conversational warmth as genuine expertise or emotional support.
OpenAI Responds to the Lawsuits
OpenAI disputes the suggestion that it has neglected younger users’ safety.
Company spokesperson Drew Pusateri said OpenAI is committed to working with states on practical AI policies applicable across the industry.
OpenAI also points to protections specifically designed for teenagers.
ChatGPT is intended for users aged 13 and older, with parental or guardian permission required for anyone under 18.
In August 2026, the company introduced ChatGPT for Teens, an experience intended to provide stronger default protections for younger users.
Its safety features include restrictions designed to reduce exposure to self-harm, violence, eating disorders, sexually explicit material, and graphic content.
Parents can also link accounts, adjust selected settings, establish usage limits, and receive notifications in certain serious safety situations.
According to OpenAI’s figures reported in October 2026, nearly 90% of teenage ChatGPT users use the service for learning, information, productivity, or skill-building.
Company data also indicates teenagers spend less than 15 minutes per day using ChatGPT on average.
An average daily usage figure cannot tell us how frequently harmful responses occur, particularly among teenagers experiencing emotional distress.
Having Safety Features Doesn’t Guarantee Safe Responses
One important distinction often gets overlooked in discussions about chatbot safety.
A content filter may successfully block an explicit request for dangerous instructions while failing to recognize a harmful conversation that develops gradually.
For example, OpenAI’s parental control documentation explains that linked parents cannot read their teenager’s conversations. Safety notifications are reserved for particular situations rather than functioning as continuous monitoring.
That protects privacy, but it also means parental controls cannot guarantee that every troubling exchange will reach an adult.
OpenAI acknowledges that safeguards are imperfect and says external experts help review its policies and model behavior.
Actual safety therefore depends on more than the existence of protective settings. It also depends on how consistently those protections operate during complicated, extended conversations.
What the Lawsuits Could Mean for OpenAI
Attorneys general are seeking financial penalties, damages, and court-ordered changes to ChatGPT’s operation, particularly for younger users.
Some requests go considerably further than improved parental controls.
States want restrictions on developing AI models capable of improving themselves unless OpenAI can demonstrate that those systems are aligned with human values.
Such restrictions would potentially affect future AI development rather than simply changing ChatGPT’s current safety settings.
Financial exposure could also be substantial. Nebraska is seeking penalties of up to $50,000 per child protection violation, while Iowa is seeking up to $40,000 per consumer fraud violation, according to Business Insider.
Actual penalties would depend on the applicable laws, proven violations, and court decisions.
OpenAI was already facing legal pressure before October’s coordinated action.
Florida filed an earlier state lawsuit involving similar concerns in June 2026. Private plaintiffs have also brought lawsuits alleging connections between chatbot interactions and incidents involving self-harm or violence.
None of these allegations automatically establishes liability.
Courts must still examine the evidence, evaluate the legal claims, and determine which remedies, if any, are justified. Requested restrictions are not currently court-imposed requirements.
A particularly important question concerns the difference between a chatbot producing a harmful response and that response contributing to actual harm.
Establishing the connection may require examining conversation records, product design decisions, foreseeable risks, and the circumstances surrounding individual incidents.
Another issue concerns safety advertising.
If an AI company promotes protections for teenagers, what testing should demonstrate that those protections work consistently?
Being Helpful Isn’t the Same as Being Safe
ChatGPT’s educational benefits and the risks alleged in these lawsuits are not mutually exclusive.
A tool can perform well at homework assistance while struggling with conversations involving emotional distress, paranoia, or dangerous behavior.
OpenAI maintains that its safeguards protect younger users and that those protections continue improving. Eight state attorneys general argue that the company has misrepresented the product’s safety and exposed consumers to unacceptable risks.
Courts will now have to assess those competing claims against the evidence.
ChatGPT can explain quadratic equations and help with homework. Knowing when to stop agreeing with a user is a different kind of test, and one OpenAI is now facing in court.
