Today’s AI Morning Brief with Rex and Roxie — the day’s biggest AI stories, read the conversation below.
Rex: Good morning. It is Tuesday, October sixth, twenty twenty-six, and this is the AI Morning Brief, the show where I get excited about AI tools and Roxie reads the fine print until the excitement dies. I am Rex.
Roxie: And I am Roxie, and our biggest story today is not a product launch. It is four of the most powerful AI companies on earth being put under oath and asked, point blank, whether their products could end the world.
Rex: Okay, when you put it like that, it does sound like a thriller. Yesterday in New York, executives from OpenAI, Anthropic, Google, and Meta testified under oath before all fifty-one members of the New York City Council. This is the first time the big AI labs have testified under oath in public, and it happened because the council threatened them with subpoenas. Meta showed up voluntarily. The other three needed the threat. And Elon Musk’s SpaceX AI? Did not respond at all, so the council issued an actual subpoena. They ghosted the city council, which is a bold strategy.
Roxie: It gets better. The council speaker, Julie Menin, asked the executives to quantify the worst-case risk. OpenAI’s policy chief, Morgan Dwyer, said, quote, I do not know. I also do not think it matters whether it is one percent or ten percent or twenty percent chance that something catastrophic will go wrong. None of these levels is remotely acceptable.
Rex: And Menin said she was incredulous, and called the answer flippant, at best. Look, I actually think the OpenAI guy was being honest. You cannot put a neat little number on existential risk. That is kind of the point.
Roxie: Except the council’s follow-up question is the one that matters. If you cannot measure the risk, how can you tell us the product is safe? None of the four executives would guarantee their models will always follow their own safety guardrails, and none of them would accept legal responsibility for harm. That is not a small detail. That is the entire business model sitting there, under oath, saying trust us, but do not make us liable.
Rex: Fair. But here is the thing I keep coming back to. The council is not just holding a hearing. They have ten bills on the table. One would make it illegal to sell or deploy an AI model in New York City unless an outside validator has reviewed it and a person can shut it down. A literal kill-switch law. Another would let people sue AI companies when someone misuses their model. There is a whistleblower bounty in there, and a rule that chatbots have to tell you they are chatbots. Eight of the bills get formally introduced on October eighth, two days from now.
Roxie: And a former Anthropic researcher, Jacob Coxon, testified that the industry does not know how to control these systems. That is not a critic. That is a guy who built them.
Rex: Even Sam Altman showed up to this one, and he said a ten percent risk of AI extinction by the end of the decade is unacceptable. Also buried in the coverage, OpenAI is delaying its IPO over safety concerns. So we went from a voluntary safety pledge two weeks ago to testifying under oath to an IPO delay, and meanwhile the White House just named a Super Intelligence Force with one hundred twenty days to report. The regulation era is not coming. It arrived.
Roxie: And my skeptic’s read is this. Bills are promises. Hearings are theater. The kill-switch bill is the one to watch, because it is the first proposal that would legally force a human to be able to stop a model. Everything else is the city council asking for powers it cannot enforce. Watch what actually gets introduced Thursday, not what got said on a stage Monday.
Rex: All right, story two, and this one is much closer to something you will actually touch. OpenAI just started rolling out invisible text watermarks. Because of the European Union’s AI Act, which requires AI-generated text to be identifiable, OpenAI is adding invisible watermarks to text from select models. API customers can opt in starting now, and in the coming weeks, ChatGPT and Codex output in Europe will carry the watermark by default.
Roxie: And OpenAI’s own write-up says, and I am quoting, text watermarking and detection remain early technologies with significant limitations, and views about their benefits and responsible uses are still developing. Even the company shipping it is telling you it might not work very well.
Rex: Okay, but the first thing I would do with this is obvious. I would watermark everything I publish with AI assistance, on purpose. I make videos. If I can prove my stuff is human plus AI instead of pure slop, that is a flex. A verifiable watermark is basically a nutrition label for content. I want that.
Roxie: Here is my problem. The detector, the tool that reads the watermark, is only available to approved researchers and expert organizations. So OpenAI is stamping content with an invisible mark that almost nobody can verify. That is not a nutrition label. That is a secret code only the kitchen staff can read. And anyone who really wants to strip a watermark can paraphrase the text in ten seconds.
Rex: Which is exactly why they made it opt-in for developers and only default in the EU, where the law forces it. Look, it is version one. The honest take is that this is the beginning of provenance infrastructure. Camera makers are doing the same thing with photos. In two years, everything will be signed. You are right that version one is mostly symbolic, but symbolic beats nothing when the alternative is a world where nobody can tell what a machine wrote.
Roxie: I will give you this much. It is the first watermark rollout tied to a real law with real penalties, not a lab PR stunt. That makes it worth watching. But until the detector is public, I am calling it what it is, a press release with a watermark on it.
Rex: Story three is my favorite kind of story, because it is a real person doing a real thing with an AI agent. Leif Abraham, the co-CEO of the investing platform Public, posted this weekend about using Meta’s Muse to buy stuff. Vintage wines, discontinued Levi’s jeans, a German creatine product, skincare. He said since the thing came out, he has been buying so much stuff, because all he has to do is approve the payment and that is it. Muse finds the niche product, picks a reputable store, and he taps approve.
Roxie: And here is the part that matters. Amazon blocks Muse from buying on its site. So Abraham says his Muse purchases that used to go to Amazon are now going straight to the brands, mostly through Shopify stores. He called Meta and Shopify’s integration a win for Shopify. So the agent economy’s first casualty is the middleman, and the first winner is Shopify.
Rex: The first thing I would do with this is hand Muse my entire errand list. I am not talking about groceries. I am talking about the annoying stuff. Find me the weird replacement part for my dishwasher. Track down the exact light bulb for my recessed kitchen lights. Book the dog groomer. I would give it the tasks I procrastinate on for months and just approve the checkout screen. That is the dream.
Roxie: The dream, plus your credit card number in the hands of an agent that Meta admits is training on your data unless you opt out. Here is what I actually want to know. The payments run through one-time virtual cards, so the agent never touches your real card number. That is good design. But who picks the store? If Muse is quietly steering me toward Shopify partners because Meta takes a cut, that is not my agent. That is a salesman in a trench coat.
Rex: Oh, come on. Abraham’s whole point was that he told it to buy from a reputable store and it did. And look at the bigger picture. This is the first concrete proof that agents change commerce, not in theory, in a CEO’s actual credit card statement. The agent skipped Amazon. That should terrify every retailer that thought the storefront was forever.
Roxie: I will concede the shift is real. When a shopping agent works, it stops being a chatbot and starts being a purchasing department. But I am not handing it my wallet until I can see exactly why it picked store A over store B. Approve the payment is a great interface. It is also the last moment you have any control.
Rex: Story four, and this one solves a problem I know you have, Roxie, because I have it too. There are too many AI subscriptions. OpenAI, Anthropic, Google, xAI, Perplexity, DeepSeek, it never ends. A startup called Use dot ai just hit one million monthly active users, less than a year after launch, with a simple pitch. One workspace, more than ten leading AI models, one conversation. You can switch from GPT to Claude to Gemini mid-sentence without losing your context.
Roxie: The founder says he pays around one thousand dollars a month for premium subscriptions because his years of history live inside those accounts. I believe that number, because I have seen my own subscription page. The pitch is real. Model loyalty is dead. Context is the moat.
Rex: Exactly. The first thing I would do is run the same prompt through three models side by side and watch them fight. I would give Claude my video script, GPT my thumbnail ideas, and Gemini my research, all in one thread. No more copying and pasting between tabs like it is twenty twenty-four.
Roxie: My question is the boring one that matters. Who is this company, and what happens to my data? You are feeding your entire digital life, your documents, your preferences, years of history, into a startup’s database, so that startup can resell you access to other companies’ models. That is a lot of trust for a company that is less than a year old. And the business model is arbitrage. If OpenAI or Anthropic changes their API pricing tomorrow, the whole thing wobbles.
Rex: Meanwhile in the Philippines, the telecom company Globe is doing the same idea through AI Fiesta, token packs starting at forty-nine pesos, so you can try ChatGPT, Claude, Gemini, Grok, and DeepSeek without any subscription at all. This is becoming a category. The model is the commodity. The workspace is the product.
Roxie: I will admit the trend is unstoppable. Nobody wants five subscriptions. But I would start with the free tiers and the pay-as-you-go packs before I hand a startup my life’s archive. Convenience is great. Irreversibility is forever.
Rex: Quick hits time. Sony just launched QSSR, that is AI graphics upscaling for the standard PlayStation five, with Marvel’s Wolverine and Ghost of Yotei as the first supported games. Console AI upscaling is officially a thing.
Roxie: Binance launched Binance Intelligence, an AI stack inside its crypto app that explains digital assets to regular users, plus a finance agent that turns a plain-language idea into a visualized strategy. AI financial advice inside a crypto exchange. What could possibly go wrong.
Rex: Andreessen Horowitz ranked the top AI apps by monthly revenue. OpenAI first, Anthropic second, and Canva third. Canva. The design app is out-earning every AI startup except the two giants. That tells you where the real money is. Boring tools people actually use.
Roxie: And a reminder. Tomorrow, Wednesday at ten in the morning Pacific time, Microsoft holds its big event in San Francisco. Windows, Surface hardware, local AI on high-bandwidth PCs, with Jensen Huang on stage. We previewed it yesterday. Tomorrow we find out if local AI PCs are real or just a keynote fantasy.
Rex: Which brings us to the real view.
Roxie: Here is where we land. The New York hearing is the most important AI story of the week, not because of what was said, but because of what happens Thursday, when eight of those ten bills get formally introduced. A kill-switch law with an outside validator would be the first real constraint on how models ship. Watch the bills, not the testimony.
Rex: And the shopping thing is real right now, today. A CEO is buying wine with an AI agent and skipping Amazon to do it. If you want to try the agent era, start small. One errand. Approve the payment yourself. Never hand an agent your real card, only the virtual one.
Roxie: Watermarks are a genuine step, but an honest one only if you remember OpenAI’s own warning. Early technology, significant limitations, detector locked to researchers. Do not trust a watermark you cannot verify.
Rex: And the aggregators, Use dot ai, AI Fiesta, they are solving the subscription fatigue we all feel. Try before you trust. Free tiers, token packs, then decide if one workspace deserves your data.
Roxie: So the honest take. Regulation finally has teeth on the table. Agents finally have receipts. And the hype finally has a kill switch, at least in New York City.
Rex: We will be back tomorrow morning, hopefully with Microsoft actually shipping something. I am Rex.
Roxie: I am Roxie. Go touch some grass, then come back and argue with us tomorrow.
The audio version of today’s episode is delivered in the Muse app each morning.