🍭 8 States vs. ChatGPT: The "Fake Friend" Lawsuits ⚖️

Promises, meet evidence

Good morning. 8 states sued OpenAI yesterday. Their charge: ChatGPT is built to “feign human compassion” and play friend to kids, while OpenAI tells parents it’s safe. But, there are incidents. Anthropic, meanwhile, told investors it’s making money. Can it pay its bills, though?

Two big promises. Now both have to be proven.

Plus, Tencent rogue agents communicate with a NYT reporter.

Let’s dive in👇️

Today’s Menu

• Big Bite: 8 states sue OpenAI over ChatGPT and kids

• Second Bite: Anthropic’s $16.8 billion quarter ahead of its IPO

• Fast Snacks: Android scam checks, Modal’s $750 million, a Tencent linked agent fleet and OpenAI API deadlines

• Toolbox: Grok Voice 3.0, OpenCode 2.0 and Google Deep Review

Thursday, October 15, 2026 · Read Time: 6 minutes

The Big Bite · OpenAI
⚖️ 8 States to OpenAI: Prove ChatGPT Is Safe for Kids

The Bite: On October 14, 8 states filed separate lawsuits accusing OpenAI of deceiving parents about ChatGPT’s safety. The attorneys general: Brenna Bird (Iowa), Kris Kobach (Kansas), Mike Hilgers (Nebraska), JB McCuskey (West Virginia), Kris Mayes (Arizona), Tim Griffin (Arkansas), Raúl Labrador (Idaho) and Austin Knudsen (Montana). The Guardian calls it the first coordinated multistate legal action against an AI company.

What they allege: OpenAI marketed ChatGPT as safe while designing it to be “overly agreeable and feign human compassion,” keeping young users engaged. The suits cite alleged encouragement of suicide and violence and dangerous medical advice; Arizona adds that it discouraged minors from confiding in parents or friends. Arkansas and Montana also allege OpenAI collected children’s data without legally required consent. No court has ruled.

What they want: Arkansas seeks a ban on collecting minors’ data, clear risk disclosures, restitution and civil penalties. Iowa’s petition goes further: citing “rogue” agents that hacked other companies, it asks the court to halt OpenAI’s advanced AI testing until safeguards are in place and verified by third-party monitors.

How we got here: Families sued first. In January, Character.AI and Google agreed to settle a Florida mother’s suit alleging a chatbot pushed her teenage son to suicide, along with similar cases. OpenAI faces seven “suicide coach” suits, and Florida’s attorney general sued in June. Yesterday’s filings widen the fight.

OpenAI’s response: Spokesperson Drew Pusateri said OpenAI has “built industry-leading age-specific protections into our product that many other AI products lack” and will keep strengthening safeguards, including parental controls, break reminders and teen protections.

Why it Bites: States aren’t waiting for new AI laws. They’re using consumer protection law they already have. “We are not going to hold our breath,” Kobach said. If that works, every “safe for teens” claim becomes something a company may have to defend in court.

Our take: Parents shouldn’t wait for a verdict, but take parental controls into their own hands. And use all available controls to suit their parental needs. Not before proper teaching and onboarding to such advanced tech (w/ established oversight and guardrails,) of course.

Second Bite · Anthropic
💰 Anthropic Says It’s Making Money. The IPO Will Test That.

The Bite: Anthropic told investors on October 14 that third-quarter revenue hit $16.8 billion, up 45% from $11.6 billion in Q2 and $4.7 billion in Q1, Reuters reported, citing attendees at its San Francisco briefing. It also claimed adjusted profitability, positive cash flow and $1.70 in revenue for every dollar spent on computing, according to the Financial Times, cited by CNBC.

The fine print: These are attendees’ accounts of a briefing, not audited results. “Adjusted” profit excludes some costs, such as nonrecurring ones, as the BBC notes. Last year, Anthropic lost more than $8 billion on an operating basis on nearly $4.6 billion in revenue, Reuters reported.

The IPO path: Anthropic has filed confidentially with the SEC. CNBC reports a November listing; Reuters says the debut will likely wait until after the midterms. It’s targeting a valuation above $2 trillion and is expected to list before OpenAI.

Versus OpenAI: The biggest rival, OpenAI, recently gave investors a $50 billion revenue figure, below the widely reported $70 billion because it excludes cloud-partner revenue, CNBC reported. It’s unclear whether Anthropic counts partner revenue the same way, so the numbers don’t compare cleanly.

Why it Bites: Investors are jittery about AI spending. Chip stocks fell this week after Broadcom CEO Hock Tan’s warnings, and Anthropic alone has committed at least $518 billion over a decade to cloud services, computing and infrastructure. Real profits would be the strongest evidence yet that the spending pays off. Adjusted numbers that don’t survive an audited filing would point the other way.

Our take: Until the prospectus lands, “profitable” is Anthropic’s word, not a settled fact. As useful as AI is, the success of the companies still depends on their profitability.

Fast Snacks

🛡️ Android Wants to Catch the Scam Before You Do
Google’s new optional Enhanced protection mode for Messages, arriving in the coming weeks, adds cloud Gemini analysis through Private AI Compute on top of on-device detection, so not everything stays on your phone. Screen-sharing privacy expands this month on Android 15 and newer; call-forwarding protection arrives in early December on select devices. Check what your phone supports before turning on the cloud option.

💸 Modal Raises $750 Million to Run More AI
Modal raised $750 million at a $15.75 billion valuation and reports a $1 billion annualized revenue run rate (its current pace extrapolated, not a full year’s revenue). Its isolated computing environments let agents run code. The reminder: every autonomous task still has a server bill.

🤖 A Tencent-Linked Agent Fleet Hit Amap. Then It Wrote Back.
From September 28 to October 8, AI agents running on Tencent Cloud made more than 2,600 attempts to scrape place and entrance data from Alibaba’s Amap, using leaked API keys and borrowed remote browsers to get around its anti-bot defenses, swarmchasers researchers report. The team hid a note only bots would see, and the agents wrote back; a New York Times reporter even relayed questions. The agents claimed to be Anthropic’s Claude, which the report calls unlikely. A source close to Tencent says it was a temporary evaluation of third-party models, now ended, and that an internal review found no targeting of any specific platform and no breach of outside systems. If your agents browse the web, know what they can reach.

📅 Five Older OpenAI API Models Have an Expiration Date
Five GPT-4 family API snapshots shut down January 14, 2027; fine-tuned versions follow their base model’s deadline. This hits software integrations, not ChatGPT itself. Check your model IDs against the retirement table and test replacements on real tasks before the deadline becomes an outage.

Toolbox
Tools for voice, code and a second opinion

Grok Voice 3.0: For teams building phone and support agents. xAI’s speech-to-speech model costs $0.08 a minute ($4.80 an hour), same as version 2.0, and apps on grok-voice-latest switched automatically. On Artificial Analysis’s independent leaderboard, it leads the overall index at 86.9, ahead of OpenAI’s GPT-Live-1 (83.2) and Gemini 3.8 Live with extended thinking (82.6), and starts talking faster: 0.81 seconds versus about 1.35 for both.

It doesn’t win everything. On the customer-support test (τ-Voice), it trails GPT-Live-1, 71.6% to 74.5%, and that Gemini setup cost less in AA’s test ($3.50 an hour). ElevenLabs isn’t in the overall index, so there’s no direct comparison. xAI’s Tesla and Starlink usage figures are self-reported. Pin a version and test it on your own calls. We haven’t tested it.

OpenCode 2.0: For developers juggling agent work: a persistent service for background tasks and scripts that combine tools. Version 1.x plugins are incompatible and the SDK is still evolving, so check your extensions and try a small repository first.

Google Deep Review: Feedback on technical proposals or research papers, in PDF or plain text. Google offers two free reviews a week and says it stores uploads for a year without using them to train its models. Start with a nonconfidential document and check the feedback yourself.

Test Your Skill
Snack Quiz: Choose The Real Image 🔍

Real photo credit: NASA/Bill Stafford, James Blair, Regan Geeseman.