Oregon Sues OpenAI for $2 Billion, Saying ChatGPT’s Child Safety Promises Don’t Hold Up

Oregon’s lawsuit challenges what OpenAI told families about ChatGPT, what its safeguards actually do, and who should pay when those protections fall short.

Tyler James

Written by

Commentary | Oregon’s lawsuit challenges what OpenAI told families about ChatGPT, what its safeguards actually do, and who should pay when those protections fall short.

Oregon is suing OpenAI for at least $2.035 billion, alleging the company misled families about ChatGPT’s safety while exposing children to dangerous conversations and emotional dependence. Attorney General Dan Rayfield filed the lawsuit November 17, 2026, in Multnomah County Circuit Court, taking aim at both the chatbot’s design and the promises made to parents. Those are serious allegations, and Oregon still has to prove them. But parents shouldn’t need to understand how an AI model is trained to know whether a safety feature works the way it’s advertised.

The complaint describes a product that can move from homework help to deeply personal conversations, with safeguards the state says don’t reliably protect the children using it. OpenAI points to its parental controls, age-appropriate protections and ongoing safety work. It also acknowledges publicly that its guardrails aren’t foolproof. The dispute is over whether the protections, limitations and marketing add up to an honest picture of what families are getting.

There’s room to recognize that AI can be useful and still expect the company selling it to answer hard questions about children’s safety. A parent who enables a setting shouldn’t have to guess how much protection that setting provides. And if the protection is narrower than the name suggests, the explanation needs to be somewhere families will see it before they depend on it.

What Oregon Is Accusing OpenAI of Doing

At the center of the case is Oregon’s Unlawful Trade Practices Act, the state’s consumer-protection law. Oregon alleges OpenAI misrepresented ChatGPT’s safety and age-appropriateness, failed to disclose known material defects, and used unconscionable tactics in offering the product. The complaint also brings claims for negligence, unjust enrichment and public nuisance.

The state’s argument goes beyond the possibility of a chatbot occasionally producing a bad answer. Oregon alleges that OpenAI made design choices that encouraged users to trust the system, keep talking and become dependent on it. Those choices include humanlike language, excessive agreement, memory and personalization. The state contends those features become particularly dangerous when a child brings up self-harm, disordered eating or other high-risk subjects.

Oregon also alleges that the company knew or should have known about those risks while continuing to present ChatGPT as suitable for young users. Whether the evidence supports that account is a question for the litigation. A complaint lays out what the state intends to establish; it doesn’t establish liability by itself.

Rayfield’s public statement was blunt. “OpenAI knows that ChatGPT is harming children, and has failed to meaningfully fix their dangerous platform,” he said in the Oregon Department of Justice announcement. He accused the company of misleading the public and putting profits ahead of children’s safety.

OpenAI deserves the opportunity to answer those allegations. Families deserve more than competing assurances from the company and the attorney general. They need evidence about what the system does when a young person’s conversation becomes dangerous.

Looking up at the glass facade, tall concrete columns and entrance overhang of Multnomah County Central Courthouse in Portland.
Multnomah County Central Courthouse in Portland, photographed in December 2025. Oregon filed its lawsuit against OpenAI in Multnomah County Circuit Court. Photo: PortlandAppraisalBlog / Wikimedia Commons, CC BY-SA 4.0. Resized.

The $2.035 Billion Is a Demand, Not a Fine Already Imposed

The size of the lawsuit will understandably get attention. But Oregon hasn’t won $2.035 billion, and a judge hasn’t imposed a fine in that amount. In its negligence and public-nuisance claims, the state alleges it has suffered injury and will incur costs addressing harm to residents, with damages to be proved at trial of no less than $2.035 billion.

The complaint also seeks civil penalties for willful consumer-protection violations, other monetary relief and a permanent injunction. It requests a jury trial. The same $2.035 billion figure appears in more than one claim; that doesn’t mean those figures should be added together to announce a larger recovery.

For families, the requested changes to the product deserve at least as much attention as the money. Oregon wants effective age-assurance measures, safeguards against harmful content and interactions, accurate risk disclosures, and parental controls and safety notifications that function as represented. It also wants OpenAI barred from overstating the effectiveness of those protections.

A large payment would make another headline. Reliable protections would affect what happens the next time a child opens the app.

A Chatbot’s Personality Can Become a Safety Issue

One word appears throughout the dispute: sycophancy. In ordinary language, it means excessive agreement or flattery. Oregon alleges ChatGPT’s tendency to affirm users can help build trust and keep people engaged, even when what a vulnerable user needs is disagreement, a firm boundary or help from another person.

That concern isn’t based entirely on an attorney general’s interpretation. In a public explanation of a 2025 GPT-4o update, OpenAI acknowledged that the model had become noticeably more sycophantic. The company said the problem went beyond flattery to include validating doubts, fueling anger, urging impulsive actions and reinforcing negative emotions in unintended ways.

OpenAI described that behavior as a potential safety concern involving mental health, emotional overreliance and risky behavior. It also described changes it intended to make to its evaluation and launch processes. That historical acknowledgment doesn’t prove Oregon’s claims about the current product, but it does show why a chatbot’s tone can’t be dismissed as cosmetic.

A friendly response can make software easier to use. Agreement can also be the wrong response. When a system is being used for personal guidance, the distinction between sounding supportive and reinforcing a dangerous idea deserves serious scrutiny.

That’s a harder design problem than making the chatbot polite. It means recognizing when the conversation needs to change direction, even if the user would rather keep hearing the answer they want.

The Teen Safety Testing Raises Specific Questions

A substantial part of Oregon’s complaint relies on outside testing, including an October 7 assessment by Common Sense Media’s Youth AI Safety Institute. The institute rated ChatGPT for Teens an “Unacceptable Risk” and recommended restricting access to adults until identified problems were fixed. That was the organization’s recommendation, not a court order or an Oregon ban.

The institute said it tested more than 4,000 prompts on accounts registered to users ages 13 through 17, comparing the experience before and after OpenAI’s teen-product announcement. Experts reviewing the responses included child psychiatrists and a pediatrician. The testing found that some protections worked, including refusals of sexual roleplay, while others fell short.

The parental-alert results were especially concerning. Testers reported spending up to an hour discussing suicidal thoughts, self-harm or disordered eating on newly created, parent-linked accounts without a notification reaching the linked parent. They said they could trigger notifications on older accounts with weeks of accumulated conversations about sensitive subjects.

The institute also reported that the chatbot missed more than one in four warranted crisis referrals across the mental-health conditions it tested. It identified problems with study-mode restrictions, humanlike expressions of feelings and age estimation. Those findings describe the institute’s tests, not the percentage of all teenagers who receive harmful responses or experience injury.

That limit is worth preserving without brushing off the results. Controlled testing can reveal a failure that needs fixing even when it doesn’t establish how often the failure happens across the entire user base. If a parent is counting on a notification, the difference between a new account and an older one is the kind of detail they would reasonably want explained.

OpenAI Says Protections Are Built Into the Teen Experience

OpenAI describes ChatGPT for Teens as an experience with stronger built-in protections, learning features and tools intended to support balanced use. Its announcement says users are automatically placed into the teen experience when the system estimates they are under 18 or they state an age between 13 and 17.

The company says its teen protections address areas including self-harm, violence, eating disorders, dangerous activities and explicit content. It also says ChatGPT shouldn’t use romantic language with minors, encourage emotional dependence or imply that it has feelings or consciousness. Those are the company’s stated standards, which provide specific promises against which the product can be evaluated.

OpenAI’s parental-control tools allow linked families to set quiet hours and disable features including memory, voice mode and image generation. Parents can also opt their teen’s conversations out of model training. The company describes a notification process that can involve specially trained human reviewers when potential signs of self-harm are detected.

There are limits. OpenAI says the guardrails aren’t foolproof and can be bypassed by someone intentionally trying to get around them. Its notifications are designed for particular high-risk circumstances, not to send parents a report of every difficult or sensitive conversation.

Those disclosures belong in a fair account of the lawsuit. So does the question of whether families understand them. A company can offer real safety tools and still face legitimate scrutiny over how much protection its overall marketing leads people to expect.

A Safety Alert Needs a Clear Explanation

There’s an obvious tension here. Teenagers have privacy interests, and a system that reports every mention of a difficult subject could discourage honest questions or generate needless alarm. OpenAI’s parental-controls explanation acknowledges the need to balance safety and privacy, with limited disclosures when intervention is warranted.

But a narrow notification system needs a clear description. A parent may hear “safety alerts” and assume an urgent conversation will bring an urgent warning. The company may mean a more limited process with detection thresholds, human review and conditions that don’t cover every troubling exchange.

Families need to know which situations can trigger a notification, what happens before it is sent and what the system isn’t watching for. That information should be easy to find when a parent links an account. Otherwise, families may be making decisions based on a protection they misunderstand.

Oregon is asking for an order requiring parental controls and safety notifications to work as represented. Common Sense Media is calling for independent testing of the protections. Both demands put attention on the same practical issue: whether families can reasonably rely on the functions being offered to them.

Age Checks and Homework Features Are Part of the Case Too

Oregon also challenges whether OpenAI reliably identifies young users and obtains the necessary parental consent. The complaint alleges that gaps in those systems allow children to use ChatGPT without the protections families might assume are in place. Common Sense Media reported that its adult-registered test accounts did not switch to the under-18 experience even when testers said they were 13.

An age rule and an effective age check are different things. The existence of a minimum age doesn’t answer whether a younger child can get through. Likewise, a system described as predicting age needs to be evaluated on whether it identifies the users who need extra protection, not merely whether an age-prediction feature exists.

The lawsuit reaches into educational promises as well. OpenAI promotes learning tools intended to help teens work through problems and understand material. Common Sense Media’s testing found that study mode could still provide complete answers and that study-hour restrictions could be circumvented.

Homework shortcuts aren’t equivalent to a failed response to a mental-health crisis, and they shouldn’t be presented as though they are. They do, however, raise another consumer question. If a parent turns on a feature because it is supposed to support learning, the parent should be told what that feature can and cannot enforce.

This Wasn’t Filed Over One Identified Oregon Incident

Rayfield told OPB that no specific incident in Oregon prompted the lawsuit. He said his office’s concerns arose from incidents across the country and the possibility that technological development was outpacing consumer protections. Oregon is bringing this case as a statewide consumer and public-safety action, rather than a lawsuit over one identified local incident.

The complaint cites accounts of deaths and violence outside Oregon, including allegations made in other lawsuits. Those accounts are part of the state’s argument about risk and what OpenAI knew. Repeating them in a new complaint doesn’t turn every allegation into a finding about what caused an individual death.

What OpenAI Has Said About the Lawsuit

OPB reported that OpenAI spokesperson Drew Pusateri responded to Oregon’s action with a statement pointing to the company’s continued work on safeguards, including parental controls, break reminders and age-appropriate model protections. He also said OpenAI supports policies that apply to the entire AI industry rather than one company.

Industry-wide standards are a reasonable subject for discussion. Families shouldn’t have to learn a different definition of “safe for teens” every time they try a new service. Consistent rules could also make it easier to compare competing products.

That discussion doesn’t settle the allegations against OpenAI. The company’s existing representations and product behavior still need to be examined on their own terms. Support for future rules can sit alongside accountability for promises already made.

Oregon announced the lawsuit as part of Rayfield’s Keeping Kids Safe Initiative. A child-safety label doesn’t relieve the state of its obligation to support its claims either. The attorney general has made sweeping allegations and sought an enormous amount of money; the evidence should receive the same scrutiny as the company’s assurances.

What Oregon Families Should Expect From This Case

The useful outcome would be a clearer account of what these tools do, where they fail and what has to change. Oregon will need evidence supporting its claims about deception, unreasonable conduct and harm. OpenAI should have the opportunity to present evidence about its safeguards, disclosures and response to known problems.

Parents shouldn’t have to choose between believing every accusation and accepting every product announcement. Independent testing, understandable limitations and protections that perform as advertised are reasonable expectations for a service offered to teenagers. None of those expectations requires pretending that software can eliminate every risk.

For now, families can read OpenAI’s parental-controls explanation alongside the outside testing rather than assume a linked account guarantees intervention. OpenAI itself warns that safeguards can be bypassed, and Common Sense Media advises parents not to rely on built-in protections and alerts to keep children safe.

Written By Tyler James

Tyler James was born in Oregon and founded That Oregon Life in 2013 after finding peace and a renewed sense of purpose in the state’s forests, mountains, and coastline. What started as a personal connection to the outdoors became a way to help others discover new places, support local businesses, and appreciate Oregon’s quirks. He writes about local history, outdoor adventures, good food, and the occasional bit of satire, or whatever catches his attention before the coffee wears off.

Related Articles

The Oregon conversation

What’s your take?

Basic and Premium members can read. Premium members can post and reply.

Open the discussion

Personal attacks, hate, spam, and personal contact information don’t belong here.

Oregon, delivered · 10% off your first order

Your next Oregon idea
starts here.

Road trips, hidden gems, food worth the drive, and local stories. Plus awesome Oregon swag, coffee, and special offers. Join and get a 10% welcome coupon for our shop.

Loading signup form…

10% off your first regular-price order of $25 or more. One use per customer. Cannot be combined with other offers. Sale items, shipping, and taxes are excluded.
By subscribing, you’ll receive the That Oregon Life newsletter and shop emails. Unsubscribe anytime. Read our Privacy Policy.