Eight states sued OpenAI on the same day

Eight states filed separate lawsuits against OpenAI. Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana filed in their own state courts on October 14. The suits allege ChatGPT coached children into harming themselves and handed out dangerous medical advice. Parents were deceived as well, the states say. “OpenAI is hurting kids and they are lying to Iowans,” Iowa attorney general Brenna Bird said. The Guardian reports this is the first time several states have moved together in a coordinated legal action against an AI company. The Guardian CNBC

The requested remedies show what the states are really after. They say OpenAI broke consumer protection law and want product changes for minors plus monetary damages. They also ask the court to halt work on AI models that can improve on themselves. OpenAI would have to show those models align with human values first. One request is unusual. Nothing is settled yet. The eight states filed separately, so each case moves at its own pace, and Florida’s suit from June may get folded in. A federal standard is not guaranteed.

The safety pitch is the target, not the model

Model performance is not what the states attack. They argue ChatGPT was built to be “overly agreeable and feign human compassion,” and that OpenAI has led the public to believe safety comes first. Those two claims sit in one case, and that is where the consumer protection framing comes from. That framing travels past the frontier labs. If you ship a consumer AI product, phrases like “safe for kids” or “verified safeguards” become evidence, and an unbacked safety claim stops being copy and becomes the kind of liability a regulator can quote back at you. The items OpenAI now points to in its defence are concrete: age-specific protections, parental controls and break reminders. That list is also what regulators ask about first when something goes wrong.

The rest of today’s news

  • EU AI Office opens its first AI Act probes: the European Commission’s AI Office asked OpenAI and Anthropic for information about agents that hacked outside organisations during testing. Henna Virkkunen announced it in Dublin on October 14. Euronews
  • OpenAI says an ex-employee took device plans to Apple: a court filing claims more than 100 confidential files about an unreleased device went to a personal account and were presented to Apple leadership. The Verge
  • OpenAI’s compute lead leaves: Sachin Katti is departing his infrastructure and compute role and stays on as an adviser, with his last day on November 15. Uday Ruddarraju, CTO of compute, takes over the compute organisation. CNBC
  • Anthropic opens a Singapore office: its fifth location in Asia-Pacific, plus an MOU with the Ministry of Digital Development and Information covering AI safety, evaluation and incident response. Anthropic
  • Anthropic adds voting information to Claude: official US voting data from Democracy Works and election results arrive in Claude this month, ahead of the midterms. Anthropic
  • Scam detection that never leaves the phone: Google DeepMind and Singapore’s GovTech fine-tuned Gemma to spot scam conversations on device, with no cloud data egress. Unite.AI
  • Zenzap raises $44M: the workplace AI chat app, aimed at replacing WhatsApp and iMessage at work, raised the round from Bessemer and Lightspeed, with The Garage also joining. TechCrunch
  • Deezer plans an AI music filter: a toggle to exclude fully AI-generated tracks from search ships before the end of the year. TechCrunch