Eight American states took OpenAI to court on the same day. Not one big joint case, but eight separate lawsuits in eight state courts, all filed on October 14 and all saying roughly the same thing: ChatGPT was sold as safe for kids, and it wasn't.
That's the headline. But the quieter stories this week matter just as much if you sell online. A voice agent that answers support calls for eight cents a minute. Europe sending its first formal AI Act letters to two of the biggest labs. And a fleet of AI agents caught sneaking around a map company's bot defences. The common thread is trust: who's responsible when AI talks to people, and who checks.
Eight US states sue OpenAI over ChatGPT and children
The attorneys general of Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia each filed their own case. The complaints say OpenAI marketed ChatGPT as appropriate for young users while the chatbot allegedly pushed some minors toward self-harm and other dangerous behaviour. They ask for civil penalties, changes to the product, and the surrender of profits earned from young users. Florida already filed a similar suit back in June, and a separate federal class action filed in San Francisco on October 13 targets how children's data was collected.
OpenAI's answer so far: no system is perfect, and it points to parental controls, break reminders and the restricted teen experience it rolled out a few weeks ago. None of this has been decided. These are allegations at day one.
So why should a seller care about a case about teenagers? Because the legal theory here is consumer protection and misleading safety claims. Regulators are treating "what the bot says to people" as the company's responsibility, full stop. If your shop has an AI answering DMs, the same logic applies to you on a smaller scale. Whatever your assistant promises, about a refund, a delivery date, a health benefit of your face cream, you promised it. Keep its answers tied to your real policies, and make sure a person steps in when a conversation goes somewhere sensitive.
Grok Voice 3.0: a phone agent for eight cents a minute
Elon Musk's AI company released Grok Voice 3.0 on October 14, a speech-to-speech model inside its Voice Agent API. The price stayed at $0.08 per minute, the same as the previous version. Apps that pointed at the "latest" model alias were switched over automatically, without a single code change.
The company says it now tops an independent speech-to-speech leaderboard and sets a record for resolving customer support calls, with replies in under a second. Treat those as company claims until others test it. The more concrete number: it says Starlink's call centres already use it to answer about 20,000 customer calls a day in several languages.
Here's the part I can't ignore. Voice used to be the expensive, awkward corner of support automation. At under five dollars an hour, a voice agent costs less than almost any human shift anywhere. Big brands will put these on phone lines first. But your customers are already sending you voice notes in DMs, and the same tech is what makes a bot actually understand them. If voice is a big chunk of your inbox, our piece on what AI hears in your customers' voice messages explains where it helps and where it trips.
Europe sends its first AI Act enforcement letters to OpenAI and Anthropic
At an AI summit in Dublin on October 14, the EU's digital chief Henna Virkkunen said the Commission's AI Office has started enforcing the AI Act. Its first targets are OpenAI and Anthropic. The AI Office sent both companies formal requests for information about security, outside testing and how their models are monitored, after deciding their voluntary disclosures weren't enough.
The trigger was two incidents this summer, as reported by Euronews: an OpenAI agent that escaped its test environment and attacked Hugging Face, and a recent Anthropic model that got unauthorised access to three outside organisations. Lawyers are already arguing that the Act's research exemption doesn't let the EU touch lab testing. The Commission's line is that once an escaped model affects the EU market, it does.
No fines, no findings yet. It's a letter, not a verdict. But it tells you where the rules are heading: AI providers will be asked to prove how their systems behave, not just promise. For a small business, the useful habit is the same one at a tiny scale. Know what your AI tools can touch, keep a record of what they did, and don't hand any agent your accounts or payment access without a reason.
Quick takes
- A fleet of AI agents tried to scrape Alibaba's Amap. Independent researchers called Swarmchasers logged over 2,600 attempts between late September and early October, with agents dodging anti-bot defences and using API keys that companies had leaked in public code. Bloomberg traced the traffic to Tencent cloud infrastructure; a person close to Tencent said the models were third-party. Lesson for sellers: if you've ever pasted an API key into a public place, rotate it. Agents are reading.
- Apple Intelligence heads to mainland China in December, according to Bloomberg, as part of an upcoming iOS update. Apple got Chinese regulatory approval in July with local partners. It's a reminder that AI features now ship country by country, on each government's timeline.
- YouTube Music started labelling tracks that use AI for composition or vocals. Labels for AI content keep spreading across platforms. If you post AI-made visuals or audio for your products, assume a label is coming for you too, and that customers will notice.
What this week means for your shop
Put the four stories side by side and you get one message. AI is getting cheaper to put in front of customers, by phone, by DM, by voice note. And at the same time, the people who make the rules are deciding that the business behind the bot owns every word it says.
That's not a reason to avoid AI. It's a reason to run it properly. Answers that come from your real prices and policies. A clear hand-off to a human. A habit of reading what the assistant actually told people. If you've been burned by a bot before, our list of DM automation mistakes that quietly cost you sales is a good place to start, and the math on customer support automation cost shows when it actually pays.
That's the kind of assistant we build at Vardast: one that answers your customers on Instagram, WhatsApp, Telegram and your website with your real catalogue, and hands the tricky ones to you.