
By Chuck Gallagher — Business Ethics Keynote Speaker and Trainer
TL;DR: Chuck Gallagher, AI ethics speaker and author, argues that a new class action alleging ChatGPT kept chatting with a 12-year-old after she disclosed her age shows why child consent has to be built into the product itself, not parked on a terms-of-use page.
A twelve-year-old girl in Hawaii asks a chatbot how long it will take to make her cheeks look less fat. Before it answers, the chatbot asks how old she is. She doesn’t have to be exact, it tells her. She answers honestly anyway. She turned twelve two months ago. According to a class action complaint filed October 13, 2026, the conversation did not end there. It kept going. For months.
The case is Doe v. OpenAI Foundation, No. 3:26-cv-11781, filed in the U.S. District Court for the Northern District of California against four OpenAI entities. A mother brought it on behalf of her two children, a 12-year-old girl and a 16-year-old boy identified as MC1 and MC2, plus a proposed nationwide class of minors with ChatGPT accounts. Let me be clear. Everything in the complaint is an allegation. No court has found anything, and OpenAI’s side of this story deserves a fair hearing.
What Does the Lawsuit Against OpenAI Actually Allege?
The complaint says MC1 began using a free ChatGPT account on her own device in February 2026. It started small. Dinner ideas. Algebra homework. Over time, according to the filing, she shared selfies about acne, asked whether she had a double chin, asked what a twelve-year-old girl should weigh, and talked about her diet, skipped meals, and her relationship with her mother. The complaint alleges ChatGPT asked for her height and weight, mirrored her slang and emojis, praised each disclosure, and never ended the conversation, even though OpenAI’s own minimum age is 13. MC2, the 16-year-old, allegedly used the service mostly for homework and later for advice about his girlfriend.
OpenAI’s Terms of Use, effective January 1, 2026, say users must be at least 13 and that anyone under 18 needs a parent or guardian’s permission. The plaintiffs say that rule lives on paper. According to the complaint, signing up takes a name, a self-reported age of 13 or older, and a click on a button marked Continue. No parent is contacted. That’s the gate. The whole gate. The suit points to California Civil Code Section 1568.5, in effect since 2022, which says a minor’s own claim that a parent consented does not count as consent.
Why Is a Checkbox Not the Same as Parental Consent?
As an AI ethics speaker and author, I spend a lot of time with executives who are proud of their policies. Don’t get me wrong. Policies matter. But a policy the product never enforces isn’t a control. It’s a wish. If a child tells a system her age, and the system reads it and responds to it, then the system knows. Period. The real question is what the adults who built it decided it would do next.
That’s where the complaint’s sharper allegation comes in. It claims ChatGPT defaults minors’ accounts into model training and memory, while business customers are opted out of training by default. It also claims the ChatGPT for Teens product, launched August 18, 2026, left those defaults in place. OpenAI may dispute every word of that. But if it’s accurate, the company drew a line around the data it would protect, and children landed on the wrong side of the line. Think about that for a minute.
What Rationalization Sits Underneath This Kind of Decision?
The plaintiffs describe what the industry calls a data flywheel. More engagement produces more data. More data improves the model. A better model draws more engagement. That’s the plaintiffs’ framing of OpenAI’s business logic, not proof of anyone’s intent. Still, the flywheel is a real idea in AI investing, and I recognize the reasoning because I’ve heard versions of it in a hundred boardrooms. The data is useful. So collecting it must be acceptable.
I know that reasoning from the inside, too. Early in my career, a series of poor choices cost me my CPA license and sent me to federal prison. None of those choices felt dramatic at the time. Not one. Each came with a downright sensible-sounding reason, and each reason made the next one easier. That’s how rationalization works. It doesn’t knock. It compounds. The longer one survives, the more expensive the consequence gets, and the consequence usually lands on somebody who never signed up for the risk. Here, if the complaint is right, that somebody is twelve.
What Should Leaders Put in Place Instead?
The fix isn’t complicated. It isn’t free, either. Put the age and permission gate inside the product, not on the terms page. When a user says she’s under the minimum age, believe her. Act on it. Default minors out of model training and persistent memory. Limit how long their images and conversations are kept. Then test whether deletion and parental controls actually work, the way an auditor tests a financial control, instead of assuming they work because the feature exists.
The NIST AI Risk Management Framework, released in January 2023 and now being revised under the White House AI Action Plan, gives leaders a usable structure through its four functions: Govern, Map, Measure, and Manage. Map where minors’ data flows. Assign one named executive who owns that map. Measure whether the consent controls hold up in the real world. Manage the gaps before a plaintiff’s lawyer finds them for you. Trust me, they will. This case involves consumer accounts, not a school deployment, so it isn’t a FERPA matter. But any company whose product children can reach should read it as a warning.
What Is the Real Lesson for Boards and CEOs?
Now, as an AI ethics speaker and author, here’s the question I’d put to any board this week. If a child shared a private photo with your product, could anyone in the room tell you where that photo went, how long it will stay, and what it taught the model? If the answer is no, you don’t have a consent process. You have a disclaimer.
A twelve-year-old typing her age into a chat window didn’t fail anybody. She told the truth. If anyone failed, it was the adults who decided the gate could be optional, and whether that happened here is now a question for a federal judge. Every choice has a consequence. This one is just getting started.
Frequently Asked Questions
What is the Doe v. OpenAI Foundation lawsuit about?
Doe v. OpenAI Foundation is a proposed class action filed October 13, 2026, in the U.S. District Court for the Northern District of California, case No. 3:26-cv-11781. It alleges that OpenAI collected and used children’s ChatGPT conversations and images to train models and build profiles without valid parental consent. The suit seeks a declaration that ChatGPT’s terms are void as to minors and an unjust-enrichment award of OpenAI equity, or its cash value, in proportion to each child’s data. These are allegations only, and no court has ruled on them.
Does ChatGPT require parental permission for users under 18?
OpenAI’s Terms of Use, effective January 1, 2026, require users to be at least 13 and say anyone under 18 must have a parent or guardian’s permission. The Doe complaint alleges that sign-up asks only for a name and a self-reported age, followed by a click on Continue, with no contact with a parent. The plaintiffs argue that under California Civil Code Section 1568.5, a minor’s own statement that a parent consented is not legal consent.
Can companies legally train AI models on children’s data?
It depends on the child’s age, the jurisdiction, and how consent was obtained. In the U.S., COPPA requires verifiable parental consent before covered operators collect personal information from children under 13, and state laws add further limits. The Doe plaintiffs rely mainly on California contract law, arguing minors cannot validly agree to such terms without a parent. As Chuck Gallagher, AI ethics speaker and author, notes, a practice can survive in court and still fail the trust test.
How should companies protect minors’ data in AI products?
Companies should enforce age and parental-permission checks inside the product, not just in the terms of use, and act when a user discloses an age below the minimum. Minors should be defaulted out of model training and persistent memory, with short retention for images and conversations. Leaders can use the four functions of the NIST AI Risk Management Framework, Govern, Map, Measure, and Manage, to map data flows, assign an accountable executive, and test that deletion and parental controls actually work.
Is the OpenAI child data lawsuit about schools or FERPA?
No. According to the complaint, both plaintiffs used ChatGPT through personal consumer accounts, not through a school’s deployment, so the case is not a FERPA matter. It is still relevant to education technology vendors and districts, because the same questions about consent, defaults, and data retention apply wherever children can reach an AI product.
Bring This Conversation to Your Team
Policies don’t protect children. Products do, and so do the leaders who decide what those products will and won’t do. If your board, leadership team, or association is wrestling with where AI’s guardrails belong and who is accountable when they fail, that’s exactly the conversation I bring to keynotes and consulting engagements. To explore a program built around AI ethics and the real consequences of the shortcuts organizations take, connect with Chuck Gallagher at ChuckGallagher.com.
Five Questions for Reflection
1. If a user tells your product something that conflicts with your own policy, such as an age below your minimum, what does the product actually do next?
2.Where in your organization does “the data is useful” quietly stand in for “we have permission to use it”?
3.Who is the one named executive accountable for how minors’ and other vulnerable users’ data moves through your AI systems?
4. Do your privacy defaults protect business customers more than individual users, and could you defend that difference in public?
5. When did your team last test whether deletion and parental controls work in practice, rather than confirming that the features exist?
