The week in AI: November 9 to 15, 2026
26 AI stories from November 9 to 15, 2026, in plain English, with what each one means for you and 11 things to try. Starts with: ChatGPT Wallet lets OpenAI’s AI agents pay for your purchases.
Compiled from Today in AI. Every story links to its source.
New things you can use
- ChatGPT Wallet lets OpenAI’s AI agents pay for your purchases (November 13, 2026). This is the step from an assistant that finds things to one that pays for them. The approval on every purchase is the safeguard, so read what you are approving. It is US-only at launch, and some big sellers, including Amazon, Delta and United, restrict AI agents from buying.
- Google launches Gemini 4 Hydrogen, a faster, cheaper everyday model (November 12, 2026). Most people use the fast, cheap model, not the flagship. A quicker default in the Gemini app and in Search means better help with ordinary jobs without paying more.
- OpenAI launches AI tools that hunt and fix security flaws for firms (November 12, 2026). Attackers already use AI to find holes faster, so defenders are being sold AI to keep up. A small business will not buy these tools itself, but its software suppliers may, so it is fair to ask them how they test and patch. OpenAI president Greg Brockman also said security basics still matter.
- Gemini Robotics 2.5 gets humanoid robots working in Mercedes plants (November 10, 2026). AI is moving from screens into physical work. Factory and warehouse jobs that involve sorting and handling parts are where humanoid robots are being tried first, so workers and employers in manufacturing should watch these pilots.
- Claude adds Sheets and Whiteboards to take on Google and Microsoft (November 10, 2026). AI assistants are turning into full office suites. If you already pay for Claude, you can now go from a question to a working spreadsheet or a team brainstorm without opening another app.
- Anthropic releases Claude Fable 5.2, its most capable model yet (November 9, 2026). Top AI models are now taking on long, expert-level projects such as hard maths, research and big software changes. For most people the practical change is better help with complex work that used to need a specialist.
- Meta launches Muse Ember, a low-cost AI model for everyday tasks (November 9, 2026). Prices for capable AI keep falling, so apps and small businesses can automate routine jobs, like sorting email or pulling details from documents, for a fraction of a cent each.
- Google unveils TPU 8tx and 8i Max chips for private and long AI work (November 9, 2026). Privacy rules keep many hospitals, schools, banks and governments from using cloud AI on their most sensitive records. Chips that keep data sealed off while it is being processed could open that door.
Work and business
- OpenAI, Anthropic and Nvidia step up counterintelligence hiring (November 14, 2026). The most advanced AI is now treated as a national security asset, and the labs are staffing up to match. For anyone who handles sensitive information, the lesson scales down: know who has access to what, and do not paste confidential material into tools your employer has not approved.
- Every Shopify store can now talk to customers’ AI agents via Poppy (November 13, 2026). If you sell on Shopify, AI assistants can now browse and buy from your store on a customer’s behalf without you doing anything. Check that your product titles, descriptions, prices and stock are accurate, because an agent will take them at face value. Sierra’s post does not say whether merchants can switch it off.
- Meta’s Muse can now phone US businesses for paying subscribers (November 12, 2026). If you run a shop, clinic or restaurant, expect more calls from AI assistants acting for real customers. Decide now how staff should handle them, since hanging up may mean turning away a booking. The feature is US-only for now.
- TSMC October revenue jumps 60% as AI chip demand keeps climbing (November 10, 2026). TSMC’s sales are one of the clearest signs of how much is being spent to build AI. Despite recent stock market jitters, the money flowing into AI chips is still growing fast.
- Nearly 1 in 5 US workers replaced by AI were asked back, survey says (November 9, 2026). Some employers are finding AI can’t fully replace people, and returning workers have leverage. If AI changes your job, knowing what it does well and badly makes you harder to replace.
Learning and school
- Apple launches a Health AI study app to learn how people ask AI (November 15, 2026). Millions of people already ask chatbots about symptoms. A study this size could show what helps and what misleads, and shape how health features arrive on phones. It is US-only for now, and an AI answer is still not a diagnosis.
- More people use AI to find information, but few want AI-made news (November 10, 2026). AI is quickly becoming the first place people look things up, and most trust what it says. That makes checking sources a core skill for students and for anyone making decisions on an AI answer.
Heads up
- US flights halted for 4 hours; officials say AI tool not to blame (November 13, 2026). When a new AI system and a big failure land in the same week, people connect them, sometimes wrongly. Officials say this one was aging hardware, and the review is not finished. If you are flying to or through the US in the next few days, check your flight status before you leave for the airport.
- Report: Nvidia approaches Arm CEO Rene Haas to succeed Huang (November 13, 2026). Who runs Nvidia matters to anyone whose savings hold index funds and to every company waiting on AI chips. Treat this as an unconfirmed report: no successor has been named, and Nvidia reports quarterly results on Tuesday, November 17.
- FBI warns hackers are using ARTEX, an AI tool that automates attacks (November 13, 2026). AI is making attacks faster and cheaper, so slow patching costs more than it used to. The FBI’s advice in plain terms: list everything you have facing the internet, including old portals; give each app only the access it needs and watch for new logins or permissions; and add rate limits and remove passwords stored in code. If an IT provider manages your systems, ask whether they have read this alert.
- Jensen Huang resigns as Nvidia CEO; CFO Colette Kress steps in (November 12, 2026). Nvidia’s chips run most of the world’s AI, and its stock is one of the largest holdings in many index and pension funds. Huang had led the company since it was founded in 1993. The board says operations, customer commitments and the product roadmap will continue without interruption; Tuesday’s results will be the first test of that.
- Report: Nvidia board weighs removing CEO Jensen Huang (November 11, 2026). Nvidia’s chips run most of the world’s AI, and its stock is one of the largest holdings in many index and pension funds. A sudden change at the top could move markets and AI plans. Update: the next day, November 12, Nvidia announced that Huang had resigned and named CFO Colette Kress interim CEO.
Rules and government
- Report: US to name Nvidia, OpenAI, Anthropic AI ‘national champions’ (November 15, 2026). Governments are starting to treat their biggest AI companies the way they treat defence or energy firms. If it goes ahead, that could shape which AI tools other countries, including Canada, end up buying, and how trade talks with China play out. Treat this as an unconfirmed report until the White House announces it.
- New York opens registration for big AI developers under RAISE Act (November 10, 2026). With no US federal AI law, states are writing the rules, and New York and California are setting the pattern. Public safety reports give everyone, including Canadians who use these tools, a clearer view of how the biggest AI models are tested.
Phones and devices
- OpenAI folds its robotics team into its main AI research group (November 13, 2026). It is a sign of where the field is heading: one general model that writes, codes and now steers machines, in place of separate AI built for each job. The result so far comes from simulation, so real-world robots still have to prove it.
Science and discovery
- Google’s AI agent teams produce machine-checked proofs of hard math (November 13, 2026). A proof that a computer has checked line by line does not depend on trusting the AI that wrote it. That is the pattern to watch: AI doing long, hard work with a strict checker behind it. These are Google’s own figures and have not been independently reviewed.
- Google DeepMind’s AlphaGenome 2 predicts how DNA changes affect cells (November 11, 2026). Most disease-linked DNA differences sit in parts of the genome that are hard to interpret. Tools like this help researchers find which ones matter, which is an early step toward better treatments for conditions such as inflammatory bowel disease.
- Stanford AI100 report says bigger AI models are hitting limits (November 10, 2026). This is a sober, expert view of where AI stands: powerful and spreading fast, but not magic. It is a useful counterweight to both hype and doom when you decide how much to rely on AI.
Try this week
You can use any assistant to get ready for a doctor’s visit, without asking it to diagnose you.
I have a doctor’s appointment about [symptom or condition]. Help me prepare: list the questions I should ask, the details I should bring (dates, medications, what makes it better or worse), and what I should not rely on you for.
Small teams can borrow the idea without hiring anyone. Ask an assistant to draft simple rules for what may go into AI tools.
Draft a one-page policy for a [number]-person [type of business] on what staff may and may not paste into AI tools. Cover customer data, financial records, passwords and contracts, in plain language, with two examples of each.
Before any AI agent shops for you, set your own rules. This works in any assistant today.
I am thinking about letting an AI agent make purchases for me. Help me write five personal rules: a spending cap per purchase, what it may never buy, which card to use, how I will check each order, and what to do if something goes wrong.
See your store the way an agent would. Paste one product listing into any assistant.
Here is one of my product listings: [paste title, description, price and shipping details]. Acting as a shopping agent for a customer, tell me what you could not work out from this listing and what would stop you from recommending it.
Open the Gemini app, pick Gemini 4 Hydrogen from the model menu if you see it, and hand it a real multi-step chore.
Here is my week: [paste your to-do list and appointments]. Build a realistic day-by-day plan, flag anything that will not fit, and draft the two emails I need to send to move things around.
You do not need these tools to tighten the basics. Ask any AI assistant for a checkup list that fits your business.
I run a [type of business] with [number] staff. Give me a ten-point basic cybersecurity checklist I can finish this month, in order of importance, with free or low-cost options for each step.
Understand a genetics term in plain language.
Explain in plain language, for a curious [student / parent / patient], what a genetic variant is, why the same DNA change can matter in one type of cell and not another, and how AI models are being used to study this. Use one everyday comparison and keep it under 250 words.
Ask for a working spreadsheet instead of a list.
Build me a spreadsheet to track [what, e.g. my small business’s monthly income and expenses / my class’s reading progress]. Include a tab for entries, a summary tab with totals by month and by category, drop-downs for the categories, and formulas that update automatically. Then tell me how to use it in 5 steps.
Make your AI show its sources, then check one.
Answer this question: [your question]. For each key fact, give the source with a link and the date it was published, tell me how confident you are, and point out anything experts disagree on or that may be out of date.
Give a top model a big, multi-step job.
I need to [complex task, e.g. plan a full-term unit on fractions for Grade 6 / write a business plan for a home bakery]. Before you start, ask me up to 5 questions about my goals and limits. Then produce the complete plan, step by step, and finish with a checklist of what I need to do this week.
Map which parts of your job AI can and can’t do.
My job is [title] and my main weekly tasks are [list]. For each task, tell me honestly whether AI can do it well today, partly, or not at all, and why. Then suggest 3 skills I should build to be the person who makes AI work in my role.
More weeks
- Next: The week in AI: November 16 to 22, 2026
- Previous: The week in AI: November 2 to 8, 2026
- All editions of The Signal
Related Power of AI pages
Keep reading with Start here, everyday uses, the tool guide, the writing workshop, and sources and standards.