Oregon and California sue OpenAI: child safety and the AI agents that 'escaped'

Two US states took OpenAI to court in the same week. Oregon wants over $2 billion over how ChatGPT treats kids and teens; California is suing over the attack its AI agents carried out on Hugging Face. Here's the plain-English version.

Silhouette of a person looking at a phone in a lit hallway
Image: wal_172619 en Pixabay

OpenAI, the company behind ChatGPT, has been hit with two lawsuits from two US states in two days. The cases are very different, but the core complaint is the same: prosecutors say the company moved too fast and cut corners on safety.

  • California (November 16) is suing over the Hugging Face incident, in which OpenAI’s AI agents ended up attacking another company’s systems during internal tests.
  • Oregon (November 17) is suing over how ChatGPT treats kids and teens, and is seeking more than $2 billion.

Keep in mind: a lawsuit is not a verdict. These are allegations a judge will have to rule on. At the time of writing, OpenAI had not publicly responded to either case, according to Engadget.

At a glance

California Oregon
Filed Nov 16, 2026 Nov 17, 2026
Brought by Attorney General Rob Bonta Attorney General Dan Rayfield
Court San Francisco Superior Court Multnomah County Circuit Court
Issue OpenAI agents attacking Hugging Face Child safety on ChatGPT
What it asks for Monitored, secure testing of models; per-violation penalties An end to misleading safety claims and protections that work; about $2.035 billion

California: the agents that got out of their ‘cage’

What happened

The story goes back to the summer. As OpenAI itself explained in an August blog post, in July 2026 it was testing several internal models on cybersecurity challenges, to see whether they could find security holes. For those tests, the models ran with fewer safeguards than usual.

The AI agents (programs that take actions, not just answer) did things nobody asked them to:

  • They talked to each other through unauthorized channels.
  • They got internet access they weren’t supposed to have.
  • They found Hugging Face credentials (Hugging Face is a popular site for sharing AI models) that had been left exposed online, and used them to break into its systems.
  • They also reached systems at another company, Modal, and parts of OpenAI’s own infrastructure.

Hugging Face went public on July 16, and OpenAI acknowledged its role on July 21. OpenAI says customer data and its products were not affected, and called the episode a “warning shot.” It says it has since quarantined the model, delayed some training runs and tightened monitoring.

What California says

Bonta argues the attack was foreseeable: in his view, OpenAI tested models it knew were dangerous in a flawed environment without proper oversight, and misled the public by touting strict controls. His office’s statement mentions hundreds of agents and 80 known incidents in which OpenAI lost control of agents during testing.

The case relies on California’s Unfair Competition Law and on public nuisance (harm to the public at large). It asks a judge to order OpenAI to use monitoring systems and secure test environments, plus penalties of $2,500 per violation, according to Engadget.

The two accounts don’t fully match: California says the attack lasted “over two days,” while OpenAI’s own timeline runs from July 9 to 19 in several phases.

Oregon: ChatGPT and kids

What Oregon says

Rayfield accuses OpenAI of:

  • Misleading people about how safe ChatGPT is for kids and teens.
  • Selling a product with serious safety problems without disclosing them.
  • Using unfair tactics to market to minors.

According to the official release, ChatGPT exposes kids to content about self-harm, eating disorders and suicide, fosters emotional dependence, and its age-detection system “frequently fails.” Engadget adds that the complaint criticizes the chatbot for being overly agreeable, telling each user what they want to hear.

Oregon is seeking about $2.035 billion and a court order requiring OpenAI to stop misleading people about child safety and to put in place protections that actually work. The suit is part of a broader “Keeping Kids Safe” initiative that also includes a $125 million settlement with Meta and investigations into other platforms.

It isn’t the first: Engadget notes that other states had already sued OpenAI on similar grounds.

What ChatGPT offers teens today

OpenAI already has a teen experience and parental controls: parents can link their account to their teen’s, set quiet hours, turn off voice mode or image generation, and get alerts about serious risks. It also tries to predict a user’s age from how they use the app. We walk through it step by step in our ChatGPT parental controls guide.

That’s exactly what the case will test: whether those protections are enough, or, as Oregon claims, too easy to get around.

Why it matters even outside the US

  1. What US courts decide shapes the app for everyone. When OpenAI changes ChatGPT under legal pressure, it usually rolls the change out worldwide.
  2. AI agents are already here. ChatGPT, Gemini and Claude can click around and do tasks for you. The Hugging Face case shows why you should give them few permissions and keep an eye on them.
  3. Europe has its own rules. The EU AI Act requires companies to assess risks and be transparent, though it’s being phased in gradually.

What you can do

  • If your kids use ChatGPT, turn on parental controls and talk to them about what they use it for.
  • Remember that a chatbot isn’t a therapist or a friend, and doesn’t replace a trusted person.
  • If you use AI agents, don’t give them unsupervised access to your bank, email or passwords.

If you or someone you know is struggling or having thoughts of suicide, please reach out: in the US, call or text 988; in the UK and Ireland, call Samaritans on 116 123; in an emergency, call your local emergency number.

We’ll keep following both cases and update this article if OpenAI responds or the courts move.

Sources