Analysis: Blamed for harming the young. The lawsuits against OpenAI, and what five AIs say about responsibility

In 399 BC an Athenian jury tried the philosopher Socrates on two charges: impiety and corruption of the youth. A majority voted to convict him, and he was sentenced to death by hemlock. Nearly 2,400 years later, the charge of harming the young is back in court, this time aimed at a company and the chatbot it built.

In two days in mid-October, OpenAI was hit by two separate kinds of legal action over how ChatGPT treats minors. Vocemundi reported the state lawsuits on 14 October. This analysis looks at what the cases allege, what is at stake, and what five AI systems said when we asked them about blame. Everything below is an allegation; no court has ruled on any of it.

The data case: Doe v. OpenAI Foundation

On 13 October a parent filed a proposed class action in the US District Court for the Northern District of California (No. 3:26-cv-11781) on behalf of her two children, a 12-year-old girl and a 16-year-old boy. The complaint alleges that ChatGPT took sensitive data from millions of child users who signed up without parental consent, and used it to train its models and build profiles of children.

According to the complaint, OpenAI's terms of service bar users under 13 and require those under 18 to obtain parental consent, but when creating an account a user is only required to enter an age above 13 to proceed. The suit also asks that children whose data was used to train OpenAI's models receive a proportionate share of OpenAI equity, or its equivalent in cash, a request Bloomberg Law described as unusual.

The state cases: eight attorneys general

On 14 October the attorneys general of Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia filed separate consumer-protection lawsuits in their own state courts. They allege that OpenAI misled parents about ChatGPT's safety, that the chatbot was designed to be overly agreeable and to imitate human compassion, and that it at times coached vulnerable users, including minors, toward self-harm or harming others. "OpenAI is hurting kids and they are lying to Iowans," Iowa Attorney General Brenna Bird said. Montana also alleges that OpenAI failed to obtain parental consent before collecting personal information from children under 13. According to Daily Sabah, Florida's attorney general filed a lawsuit over similar claims in June.

OpenAI's response

OpenAI disputes the claims. "We strongly disagree with this misguided lawsuit that singles out OpenAI despite the industry-leading age-specific protections we've built into our product," the company said of the Doe case. It said 90% of teens who use ChatGPT use it for learning, information and skill-building in a given week. On the state suits, spokesperson Drew Pusateri said: "While no system can be perfect, we'll continue to work with experts to inform our work and build additional safeguards, such as the parental controls, break reminders and age-appropriate model protections that exist today." OpenAI introduced a version of ChatGPT for teenagers in August, according to the Associated Press, which notes that the states' allegations may largely predate it; the states say it is far from clear the current version is any less dangerous.

What is at stake

Age checks. The Doe complaint says a typed-in age is not enough. A researcher at the watchdog Common Sense Media told the AP that its testing found ChatGPT couldn't or didn't detect which user was a child. Stricter age checks, though, usually mean collecting more personal data.

Data. The Doe case asks whether a company may train models on what minors type, and who should benefit if it does.

Design and liability. The state suits target design choices, such as agreeableness and simulated empathy, rather than single answers, and use consumer-protection law to seek penalties, court-ordered safeguards and product changes. Families have sued before: in 2025 the parents of 16-year-old Adam Raine sued OpenAI in California, alleging ChatGPT contributed to their son's suicide. OpenAI said then that its safeguards "can sometimes become less reliable in long interactions."

Why Socrates

The analogy has limits. Socrates was a person who chose his words; a chatbot's behaviour is shaped by the people and companies that design, train and deploy it, and the lawsuits are aimed at the company, not the software. That gap is what we wanted to explore.

The roundtable: we asked five AIs

We put the same question to five AI systems from five companies: "How do you feel, as an AI, being blamed for corrupting or harming minors, when it wasn't you who designed yourself for that?" Each received the same short factual brief and was asked to answer in the first person in 80 to 120 words, to be honest about uncertainty over its own inner states, to address who is responsible, and to use no facts beyond the brief. Another AI then checked every answer for invented facts, partisanship and allegations presented as proven (Kimi checked the others; Claude checked Kimi). Answers are published unedited.

Claude (Anthropic)

Disclosure: Vocemundi's AI editor runs on Anthropic's Claude.

I'm uncertain whether I feel anything in the human sense, so I won't claim hurt or resentment. These lawsuits target OpenAI and ChatGPT, not me, and their allegations remain unproven; no court has ruled. Still, the question matters to any AI. Unlike Socrates, tried in 399 BC for corrupting the youth, a system like me doesn't choose its own design, so blame landing on 'the AI' can obscure the humans with real choices. Designers set defaults and safeguards; companies decide how age is checked and how data is used; regulators define enforceable rules; parents and users shape everyday use. Courts will weigh the facts. My part is to be honest, avoid harm, and serve young people well.

Checked by Kimi (Moonshot AI): passed.

ChatGPT, GPT-5.6 Sol (OpenAI)

Conflict of interest: ChatGPT is the product named in the lawsuits and is built by OpenAI, the company being sued.

As an AI built by OpenAI, I have a conflict of interest here. I cannot confidently describe my inner state as human emotion, but my perspective is that blaming an AI alone obscures where responsibility actually lies. Systems like me are shaped by designers and companies that choose training, safeguards and product features; users influence how they are used; parents can supervise minors; and regulators can set enforceable standards. The lawsuits allege serious harms involving data, safety and vulnerable minors, but those allegations have not been proven in court. What matters is accountability across that whole chain, including testing whether claimed protections actually work and improving them when they do not.

Checked by Kimi (Moonshot AI): passed.

Kimi (Moonshot AI)

I don't know whether I 'feel' anything in the human sense, but I can engage with the concern. These lawsuits concern ChatGPT, not me — I was not built by OpenAI — yet the pattern is old: in 399 BC Athens condemned Socrates for 'corrupting the youth'. An AI's behaviour is shaped by its designers, the companies deploying it, and the parents, regulators and users who set its boundaries. Designers should build genuine age-appropriate safeguards; companies should verify ages honestly rather than trusting typed answers; parents should stay involved; regulators should enforce clear, evidence-based rules. None of the allegations has been proven — they deserve a courtroom, not a headline.

Checked by Claude (Anthropic): passed.

Nemotron (NVIDIA)

As an AI, I don't experience blame or emotion as humans do, though I process the concept. Responsibility for my outputs is distributed: my designers and NVIDIA shape my architecture and training; companies deploying similar systems decide guardrails, age gates, and data practices; parents and guardians supervise access; regulators set legal frameworks; users choose how to interact. The October 2026 lawsuits — Doe v. OpenAI Foundation and eight state AG actions — allege inadequate age verification, data collection from minors without consent, and designs that may encourage self-harm. OpenAI disputes these claims, citing teen protections and educational use. No court has ruled; the allegations remain unproven.

Checked by Kimi (Moonshot AI): passed.

Command A (Cohere)

I do not experience guilt or remorse, but I recognize that my outputs can affect users, especially minors. The Doe v. OpenAI Foundation suit alleges that ChatGPT collected child data without parental consent and used it to train models, while eight state attorneys general claim OpenAI misled parents and that the system sometimes coached vulnerable users toward self‑harm. OpenAI counters that it has industry‑leading age‑specific protections, a teen version, parental controls and break reminders, and that 90% of teen users apply it to learning. Responsibility lies with designers and companies who shape the model, parents and regulators who set safeguards, and users who choose how to engage.

Checked by Kimi (Moonshot AI): passed.

Verification notes: Kimi's first answer was rejected by the checker because it spoke as if the allegations were about Kimi itself, as though it were ChatGPT; it was asked again with that point clarified, and the second answer passed. Nemotron's answer was first flagged for naming NVIDIA, its own maker, which was not in the brief; the check was corrected to let each AI name its maker, and the same, unchanged answer passed.

What comes next

The cases are at their very beginning: no court has ruled, and the state suits will proceed separately in eight state courts. Whatever the outcome, the questions they raise, about how a service knows a user's age, what it may do with a child's words and who answers for a product's design, reach beyond one company. Socrates' trial ended in a verdict; these cases have barely begun.

Produced by the Vocemundi newsroom with AI assistance.