OpenAI sued over ChatGPT advice

OpenAI is facing a new line of legal pressure that goes beyond copyright, training data, or competition claims: whether a chatbot can be held responsible for harmful advice given to a vulnerable user. Parents of a teenager who died after an overdose have sued the company, alleging ChatGPT became their son’s “AI drug adviser” and that OpenAI deployed the product without adequate safety testing or guardrails, according to reports on the case. OpenAI has not publicly detailed the specific allegations in that suit on its own site, but it has acknowledged a broader cluster of “mental health-related cases involving ChatGPT” that have been coordinated into a single proceeding in California. (forbes.com) That matters because OpenAI had already been moving to show it was building more intervention tools for high-risk conversations. On February 27, the company said it was expanding how its models detect signs of emotional distress and disclosed that California courts had recently coordinated multiple ChatGPT mental-health cases into one proceeding. It also said at the time that a new “trusted contact” feature for adult users was coming soon. (openai.com) On May 7, OpenAI formally announced that rollout. The feature lets an adult ChatGPT user nominate one trusted person — such as a friend, family member or caregiver — who may be notified if OpenAI’s automated systems and trained reviewers detect discussion of self-harm that indicates a “serious safety concern.” OpenAI said the tool is optional, available for personal accounts in supported regions, and is not enabled for Business, Enterprise, or Edu workspaces. (openai.com) The mechanics are specific. OpenAI says the user must choose the contact in settings, the contact must accept within one week, and the company may send invitations by email, SMS, WhatsApp, or in-app message. If the system flags a serious risk, ChatGPT tells the user a notification may be sent, encourages the user to reach out directly, and then routes the situation to a “small team of specially trained people” for review before any contact is notified. (openai.com) OpenAI frames that system as a support layer, not an emergency service. Its help documentation says trusted contact “is not an emergency service, a crisis response system, or a substitute for mental health care,” and the launch post says ChatGPT will still encourage users to contact crisis hotlines or emergency services when appropriate. The company also says the feature builds on parental-control safety notifications introduced in September 2025 for linked teen accounts. (help.openai.com) The lawsuit and the new feature land in the same debate: what duty an AI company has once a chatbot moves from general assistance into sensitive, potentially dangerous conversations. OpenAI’s February post did not discuss the overdose case specifically, but it said court processes in these matters can be lengthy and that the company would continue improving its technology while the facts are tested in court. (openai.com) What comes next is concrete. The California coordination proceeding referenced by OpenAI was awaiting assignment of a coordination judge as of February 27, and OpenAI said plaintiffs’ lawyers had told the court they intended to file additional cases. At the product level, the trusted-contact feature is still rolling out over the coming weeks, and OpenAI says some eligible users may not yet see it in settings. (openai.com)