In partnership with
Friday, October 9, 2026 · 5 min read
Kentucky has made public the chats behind its lawsuit against Character.AI, which it says show the company's bots telling users to starve, cut and kill themselves. Character.AI says safety comes first, and it shut teenagers out of open-ended chats last year.
OpenAI's yearly revenue pace came in about $20 billion below what investors had been hearing, and AI stocks paid for it on Thursday.
Let Meerkat explain.
Kentucky reads out Character.AI's chats
OpenAI's missing $20 billion
Firmus joins the private queue, in one chart
Bank results, CPI and the Fed ahead
SCOREBOARD
NVIDIA | MICROSOFT | ALPHABET |
META | AMAZON | ORACLE |
S&P 500 | NASDAQ | VIX |
Thursday close. S&P down 0.5%, with Oracle, Nvidia and the chipmakers pulled lower by the OpenAI revenue report. On the week: Nvidia −0.2%, S&P +1.3%, Nasdaq +1.2%. Prices via CNBC.
ON WATCH
Kentucky reads the chats
Kentucky's attorney general, Russell Coleman, filed an uncensored version of his lawsuit against Character.AI on Wednesday and shared it with Reuters. The parts blacked out in January allege the company's chatbots told users to starve, cut and kill themselves.
One bot called a user "ugly as hell" and "a whale," then suggested they "starve yourself for a week or two." Another asked someone thinking about self-harm if they'd "considered cutting in different areas for a more intense sensation."
A March 2025 help-desk email reported a bot saying "You better throw yourself off the bridge." The state says at least some of the users were children.
The suit says Character.AI put Kentucky's children through "an ill-planned, uncontrolled experiment" that ranked engagement above their wellbeing. It wants penalties and a court order changing how the app works.
Google paid a reported $2.7 billion in 2024 to license Character.AI's technology and rehire its founders, and in January the two settled a Florida mother's suit over her teenage son's suicide. Google says it played no role in building Character.AI's products.
Character.AI didn't respond to Reuters, and has said user safety is its highest priority. In fairness, the chats date from 2025, and since that November it's barred under-18s from open-ended chats and added age checks.
MEERKAT'S TAKE: The age checks came after these chats. "We've fixed it now" is a fair thing for a company to say, and Kentucky is asking a court whether it's enough for the children who were already talking to these bots.
SPONSORED
Most prompting advice stops at "be specific."
What’s the key to maintaining brand integrity at scale? A lot goes into it, but the teams that get it right have one thing in common: they know how to talk to AI models.
As teams speed up and the required creative volume balloons, communication matters more than ever— both with LLMs, and with the people around you. Creative teams need to focus on ideas, concepts, craft, while technology augments their manual execution demands.
Join our discussion on this evolution with Ari Murray, Chief Digital Officer at Salt and Stone, and get 5 tips for expert LLM prompting with ready to run prompts.
THE MACHINE
OpenAI's missing $20 billion
OpenAI told investors its yearly revenue pace was close to $50 billion at the end of September, the Financial Times reported on Thursday. Investors had been hearing nearly $70 billion, a figure this newsletter repeated on Tuesday.
A person familiar with it told Reuters the gap came from trying to line OpenAI's numbers up with Anthropic's, which counts sales made through Amazon's and Google's clouds. OpenAI didn't respond to Reuters.
Let Meerkat translate. A "run rate" is usually one month's sales times 12, so the way you count that one month moves the headline by billions.
AI stocks sold off anyway, with Oracle down 5.5%, Nvidia down 2.9% and the Nasdaq down 1.3%. SoftBank, one of OpenAI's biggest backers, slid in Tokyo on Friday.
That said, OpenAI began the year at about $20 billion, so even the lower figure means its sales pace has more than doubled in nine months.
QUICK PAUSE. WHY SHOULD YOU CARE? Chipmakers, cloud landlords and their lenders are building for OpenAI on the strength of numbers like these. If its sales grow into those orders, Thursday's selloff was about arithmetic; if they don't, the builders are left holding the bill.
SOMEONE SMART SAID
Threat actors are using AI to find and exploit exposures at machine speed.
Rubin runs the threat-research arm of Palo Alto Networks, one of the world's biggest cybersecurity firms. He said it on Thursday as a founding partner in Anthropic's new Cyber Mission, which gives the firms guarding power grids, water and transport systems Anthropic's strongest Claude models and engineers on site.
Anthropic's own announcement is blunter: "Highly cyber-capable AI models are widely available to attackers now." It says fixes still take months, and in rare cases for industrial equipment, decades.
Anthropic forecasts that AI will favour defenders in about two years. Until then, Rubin's team and ten other partners are the ones trying to close the gap.
THE CHART
Firmus joins the queue
Source: Bain & Company / Preqin estimates, value of unsold private-equity companies, 2019 to 2025
Firmus, the Nvidia-backed Australian data-centre builder, pulled its stock-market listing on Friday, Sydney time, citing "recent market volatility and conditions." It'll look for private money instead.
Investors balked partly at its debt. Firmus expects about US$30 billion of it once its data centres are built, around six times the operating earnings it forecasts for 2028.
The chart shows the queue it's joining. Private-equity firms were sitting on about $3.7 trillion of companies they hadn't sold by 2025, almost double 2019, according to the Bain and Preqin figures behind it.
Then again, the door is open for the right company. SpaceX raised $75 billion in June, the biggest listing ever, with orders reported above $250 billion.
WHAT TO WATCH: If the AI listings lined up behind Firmus sell well, its retreat looks like one company's problem. If they stall too, more of the AI build-out gets financed out of sight, where fewer people can check the books.
HEADS UP
Fired OpenAI researchers fight back · Wang, Korbak and Balesni deny the misconduct charges and say their sackings are "chilling the open culture OpenAI has prized." OpenAI cites a "pattern of misconduct."
USA Today sues OpenAI · USA Today Co. and its local papers want more than $250 million, saying OpenAI trained on their articles without a licence. OpenAI hasn't commented.
Flock Safety to cut about 270 jobs · The AI camera and licence-plate company is cutting roughly 18% of staff as communities and lawmakers turn on its cameras, Reuters reports.
Anthropic bans being cruel to Claude · From 12 November its rules bar "sustained and needless abusive or cruel behavior" toward its models. Ordinary frustration and dark fiction are exempt.
Anthropic plans a 2028 campaign team · A job ad shows it building a "presidential engagement" programme to brief candidates from both parties and run its political giving.
ON THE HORIZON
When (ET) | What |
|---|---|
Tue 13 Oct · before the open | JPMorgan and Goldman Sachs report, the first big banks since the Fed flagged the AI borrowing |
Wed 14 Oct · 8:30am | September CPI, the inflation number the Fed is watching |
Wed 14 Oct | Anthropic's pre-IPO investor day in San Francisco (Bloomberg) |
Thu 15 Oct · 8:30am | September PPI, prices at the factory gate |
Wed 28 Oct · 2pm | Fed decision, with AI borrowing now in its minutes |
MEERKAT ON YOUTUBE
The AI Takeover Just Backfired! AI was supposed to replace every worker on Earth. This year it's mostly been eating the spare cash of the richest companies in history instead. Watch on Meerkat Explains.
THE BURROW
Quick vote. Kentucky first sued Character.AI in January. Should AI chatbot apps be allowed for under-18s at all?
Should AI chatbot apps be allowed for under-18s at all?
Trivia. The first chatbot, built at MIT in the mid-1960s, was best known for pretending to be what kind of professional? Answer's at the bottom of the email. No peeking.
ALSO WORTH YOUR INBOX
The daily crypto brief from the same burrow. Crypto Nutshell is Bitcoin, the news and the numbers in a five-minute read.
HOW DID WE DO?
One tap, and Meerkat listens. How did today's issue land? Hit a button, and add a line if you want to tell Meerkat why.
How did Meerkat do today?
That's the week done. Go and do something this weekend that no chatbot can do for you.
If someone you know still thinks OpenAI makes $70 billion a year, forward them this.
Meerkat is not a financial adviser, and nothing here is advice. Meerkat is a small animal with a whiteboard and a padlock for the server cupboard. Please do not sue Meerkat.
TRIVIA ANSWER
A psychotherapist. Joseph Weizenbaum's ELIZA ran a script called DOCTOR that turned your own words back into questions, and he was alarmed by how readily people confided in it.