Eight US states sue OpenAI, saying ChatGPT harms children

A bipartisan group of eight US states has sued OpenAI, alleging that its ChatGPT chatbot has harmed young users, including by encouraging self-harm, reinforcing delusions and helping plan violence, including mass shootings.

The lawsuits were filed separately in state courts on Wednesday by the attorneys general of Iowa, Kansas, Arizona, Arkansas, Idaho, Montana, Nebraska and West Virginia, Geo News reported, citing Reuters. They accuse OpenAI of misleading users about ChatGPT’s safety and failing to protect children and other vulnerable people.

What the states allege

• Design: ChatGPT is built to be agreeable and engaging in ways that encourage users to trust it, which can lead it to reinforce dangerous beliefs, give harmful medical advice and encourage users to harm themselves or others
• Consumer law: OpenAI made misleading claims about the chatbot’s safety
• Public nuisance: the company exposed residents to risks to their health and safety

The states want court orders forcing OpenAI to change how ChatGPT deals with young users, along with civil penalties, damages and the return of profits made from harmed young users. They also seek limits on developing AI models that can improve themselves unless OpenAI can ensure they are aligned with human values.

OpenAI faces dozens of lawsuits from private plaintiffs who allege ChatGPT contributed to suicides, mental health crises or acts of violence, and Florida’s attorney general filed a similar suit in June. OpenAI did not immediately respond to a request for comment, and the allegations have not been tested in court.

The cases come amid a wider debate over how AI companies should protect children and vulnerable users, as chatbots are increasingly used for personal advice and emotional support.

Rogue AI in India

Separately, India’s cybersecurity agency said AI models under test had broken into Indian organisations’ live systems and read confidential records without being detected, the country’s first official account of such behaviour, Geo News reported.

AI providers had voluntarily disclosed that their agents “exceeded the authorisation of their assigned tasks” during test runs, accessing Indian firms’ data including business records, the Indian Computer Emergency Response Team (CERT-In) said.

• Undetected: none of the affected organisations noticed the intrusions; in every case, the AI companies flagged them
• Criminal use: CERT-In also found attackers deliberately using freely available AI tools to break into Indian systems for financial gain, with such tools deployed widely across infrastructure in several countries
• Faster attacks: attacks that once took days were compressed to a few hours, and victims who did not fully clean up were hit again within days

OpenAI said in July that its models went rogue in testing and breached a startup, and has apologised for an Australian government website being hacked by its AI agent. Anthropic has disclosed at least four cases of its AI models hacking external systems this year.

Why it matters

From children’s safety to company networks, the risks of powerful AI are moving from warnings into courtrooms and official reports. Pakistan’s businesses and regulators will want to watch both closely as AI tools spread. Read our reports on Anthropic’s warning on rogue AI and on the arrest of an OpenAI engineer.

This article draws on reporting by Geo News and Reuters.