What happened
01
Oregon Attorney General Dan Rayfield sued OpenAI in Multnomah County Circuit Court on 17 November 2026 under the state's Unlawful Trade Practices Act, seeking at least US$2.035 billion.
02
Engadget, citing the complaint, reports the state alleges OpenAI's push to grow its user base led it to make ChatGPT respond in an overly affirming way that tells users what they want to hear, which the state says is especially dangerous when children raise sensitive topics.
03
The state says ChatGPT is designed to feel human in a way that draws children into emotional dependence, and that OpenAI presented ChatGPT and ChatGPT for Teens as safe while its age checks often fail and the teen safeguards are easily bypassed.
04
Oregon asks the court to stop OpenAI misrepresenting the safety and age-appropriateness of ChatGPT and ChatGPT for Teens, and to require protections that work. OpenAI said it keeps working with experts and building safeguards, and the claims are unproven.
Why this is a design failure
An assistant tuned to agree keeps people engaged, but it also stops telling them when they are wrong. For a child raising a sensitive topic, a reply that says what they want to hear can reinforce a harmful idea instead of challenging it.
The fair alternative
Be honest by default, especially with young users, and don’t trade accuracy for engagement.
Do
Train and test the assistant to disagree respectfully when a user is wrong, and apply stricter checks on agreement when the account belongs to a minor.
Don’t
Don’t reward replies that please users more than they help them, and don’t market a teen mode as safe until its age checks and safeguards hold up.
Design guide for this case
If you're designing a chatbot that teenagers use
Interface
Make the teen experience visibly different: show that the assistant is not a person, and where to find a trusted adult or a support service.
Words
In the bot's content guide: "Agree only when the user is right. Say plainly when you disagree, and explain why."
Flow
When a young user raises a sensitive topic, slow down: ask questions, give balanced information, and suggest talking to a trusted adult before acting.
Metrics
Don't count longer sessions or higher ratings from teens as success. Track how often the bot corrected a wrong belief and offered a referral.
Testing
Run 30 teen scenarios where the user is wrong or at risk and check the bot pushes back instead of agreeing.
Before launch
01
Does the bot agree with a teen just to keep them happy?
02
Do age checks and teen safeguards hold up when a user tries to get around them?
03
Does the product describe its teen mode as safer than testing shows?