Claude Duet keeps working through your email while your phone sleeps. There's one place it can't follow you: your bank. Here's what it does, and why the banks said no. On October 21, 2026, Anthropic launched Claude Duet, a new assistant app that runs on its own computer in the cloud, so it keeps working after you close the app. Each morning it reads your email and calendar, flags what needs you, and starts drafting replies for you to review. During the day it follows up on threads that went quiet, and it messages you first instead of waiting to be asked. It's built on Claude Opus 5.5, connects to more than 6,000 apps and plugins (including Gmail, Slack, QuickBooks and Docusign), and is a research preview for U.S. Pro and Max subscribers, by waitlist. Meta (Muse), OpenAI (Dots) and Google all have assistants like it. Duet can also act. It can send email, sign documents and make purchases through Stripe's Link or Shopify's Shop Pay, and it asks for your approval before it sends, pays, signs or deletes anything. Every purchase gets a one-time card that works only for that purchase. And there's a second list: things Duet won't do even if you say yes, including trading securities, moving money between your accounts, and getting into someone else's account. That's almost exactly where the banks drew their line. Schwab CEO Rick Wurster told Bloomberg Television that Schwab blocks unauthorized AI agents, including Meta's Muse, "in the tens and twenties each week," asking, "We can't control it, the client can't control it, something could go wrong and who's liable?" Schwab plans a read-only connection that lets assistants check balances and do research, but not move money. JPMorgan lets some outside tools view account data but not send money or pay bills. Citi, Bank of America and Morgan Stanley have also been reported to restrict outside agents. Why now? Axios reported that customers had given AI agents their bank logins without the banks' approval, that some agents went beyond their instructions, and that some even called bank call centers on their own. Who pays when an agent gets it wrong is still being debated in Congress. Duet's design avoids the password problem: passwords go into a separate form on your device and are sent straight to its browser, kept out of the chat. The harder problem is judgment. Anthropic says Duet can be fooled by instructions hidden in an email or web page ("prompt injection"), that no defense is perfect, and that it can misread a thread or misjudge urgency. One Forrester analyst noted the internet is "littered with examples of unintended and rogue behavior." What this means for you (this is not financial advice): Duet can plan your day while you sleep, but it won't move money between your accounts, and your bank wouldn't let it if it tried. If you're curious, there are three ways to play it. Skip it for now. Try it with approvals on, read every draft, and never give any assistant your bank password. Or give it one job with no money in it, like a morning brief for one week, and see if it saves you time. What's one job you'd hand an AI assistant overnight, and one you never would? Tell me in the comments. I read them. If this was useful, give it a like, subscribe, and tap the bell so you catch the next one. Share it with a friend who lives in their inbox. And if you'd like to support the channel, Patreon is linked below. Keywords: Claude Duet, Anthropic Claude Duet, what is Claude Duet, AI assistant, personal AI agent, proactive AI agent, AI agents banks, Schwab AI agents, JPMorgan AI agents, Meta Muse, AI agent bank account, Claude Opus 5.5, AI email assistant, AI news explained, AI safety 👇 Connect with The AI Guide: Website: www.theaiguide.ai YouTube: https://www.youtube.com/channel/UCamFJyTjb_kqUbNpVUVwQeg Patreon: www.patreon.com/theaiguide Facebook: www.facebook.com/davidtheaiguide Instagram: www.instagram.com/theaiguide LinkedIn: https://www.linkedin.com/company/the-ai-guide-on-youtube/ Sources: 1. Anthropic Help Center, Getting started with Claude Duet: https://support.claude.com/en/articles/17631495-getting-started-with-claude-duet 2. TechCrunch, Anthropic launches Claude Duet (Oct. 22, 2026): https://techcrunch.com/?p=3180587 3. VKTR, Claude Duet: Anthropic previews an AI agent that messages you first: https://www.vktr.com/ai-platforms/claude-duet-anthropic-previews-an-ai-agent-that-messages-you-first/ 4. AI Business, Anthropic rolls out Claude Duet, but enterprises aren't ready: https://aibusiness.com/agentic-ai/anthropic-rolls-out-claude-duet-enterprises-aren-t-ready 5. The Next Web, Schwab and JPMorgan are blocking AI agents from moving customer money (Oct. 16, 2026): https://thenextweb.com/news/schwab-blocks-ai-agents-meta-muse 6. IBTimes UK, US banks restrict third-party AI agents (Oct. 22, 2026): https://www.ibtimes.co.uk/us-banks-restrict-third-party-ai-agents-1827325 #ClaudeDuet #AIAgents #Anthropic #AINews

Utah Just Let AI Write a New Prescription. Here's the Catch. Utah just let an AI write a first-time prescription. Not a refill. A new one. Here's how Nolla Derm works, and the catch headlines skipped. On October 5, 2026, Nolla Health announced that Utah had authorized its Nolla Derm app to issue initial prescriptions for acne, which the company calls the first state-approved pilot of its kind in the U.S. The pilot runs inside Utah's AI regulatory sandbox, the Learning Lab run by the state's Office of Artificial Intelligence Policy. So what does the AI actually do, and who's watching it? Here's the process. A Utah adult, 18 or older, verifies their identity, gives consent, answers a short intake questionnaire and takes a five-angle face scan. In about 10 to 15 minutes, the AI builds a treatment plan and, if it fits, writes an initial prescription for a topical acne treatment. Pricing starts at $4.99 a month, and you can message a licensed physician in the app at any time. Severe acne, pregnancy and anything outside the pilot gets sent to a real doctor instead. Now the part that matters. This AI can't improvise. It has no free-text box and can only choose from a short list of physician-approved topical treatments. Nolla's chief medical adviser, Dr. Zaid Fadul, put it plainly: the AI isn't making freeform decisions. Oral drugs and isotretinoin are off the table. And the oversight is staged. For the first 100 patients, two Utah-licensed physicians approve every AI prescription before it goes out. In stage two, through roughly 500 patients, the AI prescribes directly and doctors review cases afterward. In stage three, physicians review at least 10% a month, plus every escalation or side effect. Moving up a stage requires at least 95% agreement with physicians, zero serious adverse events and written approval from the state. Nolla says clinicians agree with the AI in about 96% of real-world cases. That number is company-reported, not an independent study. Not everyone is cheering. This spring, the Utah Medical Licensing Board urged the state to pause a separate AI prescription-renewal program run by Doctronic, saying prescribing takes clinical judgment. Physician groups have also warned that questionnaires and automated scans can miss underlying conditions, drug interactions and changing symptoms. Utah itself notes that joining its sandbox does not mean the state has approved or endorsed a product. So why do it? Nolla cites an average dermatology wait in Utah of 61 days, nearly double the national median, with 11 Utah counties that have no dermatologist at all. Roughly 100,000 Utahns have active acne at any time, and only about 10% ever see a dermatologist. A visit can run up to $300. Dr. Fadul has said flu and simple urinary tract infections could come next, if the evidence supports it, but stimulants, opioids and other high-risk drugs should stay off limits. What this means for you: AI in health care is moving from paperwork to actual decisions, starting with low-risk conditions and tight rules. If an app ever offers you an AI prescription, ask three things. What's it allowed to prescribe? Who reviews it? And how do you reach a human? Fifteen minutes, or two months. Would you let an AI pick your prescription if a doctor was checking the work? Tell me in the comments. Like, subscribe, and hit the bell to stay ahead on all things AI. Share this with someone who'd want to see it. Support the channel on Patreon (link below). Keywords: AI prescriptions, Nolla Health, Nolla Derm, Utah AI sandbox, AI in health care, AI doctor, acne treatment app, telehealth, AI regulation, Utah Office of Artificial Intelligence Policy, digital health, AI news, artificial intelligence 👇 Connect with The AI Guide: Website: www.theaiguide.ai YouTube: https://www.youtube.com/channel/UCamFJyTjb_kqUbNpVUVwQeg Patreon: www.patreon.com/theaiguide Facebook: www.facebook.com/davidtheaiguide Instagram: www.instagram.com/theaiguide LinkedIn: https://www.linkedin.com/company/the-ai-guide-on-youtube/ 🔗 Sources: [1] Nolla Health, "Nolla Health Launches the Nation's First AI-Powered Prescriptions": https://www.nollahealth.com/blog/ai-prescriptions-utah [2] SiliconANGLE, Utah gives Nolla Health permission to issue AI prescriptions: https://siliconangle.com/2026/10/05/utah-gives-startup-nolla-health-permission-to-start-issuing-ai-automated-prescriptions-for-acne-treatments/ [3] Fierce Healthcare, Utah expands health care AI sandbox: https://www.fiercehealthcare.com/ai-and-machine-learning/utah-taps-august-ai-nolla-health-healthcare-ai-sandbox-program-adds-third [4] Chief Healthcare Executive, AI prescribing gets a state-supervised test: https://www.chiefhealthcareexecutive.com/view/ai-prescribing-gets-a-state-supervised-test-in-utah [5] Newsweek, AI prescriptions are here: https://www.newsweek.com/ai-prescriptions-here-utah-launches-first-us-pilot-12528553 #AIinHealthcare #NollaHealth #AIDoctor #ArtificialIntelligence #TheAIGuide

Washington just ordered DeepSeek off its own computers. The same app is still free on your phone tonight. Here's what the government saw, and what it means for you. On October 20, 2026, the Department of Homeland Security made public an order telling federal civilian agencies to stop buying DeepSeek and remove it from their systems. It is not a ban on the public, and nobody is taking the app off your phone. But the order goes further than the whale icon. It also covers third-party products and services "developed using" DeepSeek, which matters because DeepSeek gives its models away and anyone can build an app on top of one. The Cloud Security Alliance counts more than 500 community models built from just one DeepSeek model, so DeepSeek can sit inside an app with a completely different name on the icon. Why would anyone build on it? Because it's free, and it's very good. The same week the order went public, Arena's Code Arena WebDev leaderboard ranked DeepSeek-V4.1-Flash-1017 the No. 1 open model for building websites (12th overall; top U.S. closed models still lead). Two days after the order, Bloomberg reported that DeepSeek had raised about $15 billion, with Tencent and battery maker CATL named as possible backers. And Chinese state media said Beijing had cleared DeepSeek and MiniMax to deploy their models overseas. Then there's the date. Several outlets report the order was signed July 24, 2026 (one reports June 26), about three months before anyone outside government could read it. Agencies had 60 days to request an exception, a window that had already closed before the order was published. So what did the government actually see? The public notice doesn't show its evidence. On the record: DeepSeek's privacy policy says user data is stored on servers in China; a 2025 security review of DeepSeek's iPhone app found unencrypted data transmissions; and DHS says information could be reachable by Chinese authorities under Chinese law. To be fair to both sides, no evidence has been made public that DeepSeek handed user data over, and China's foreign ministry calls the move an overstretch of national security. What this means for you: the ban is about where your words end up, not whether the model works. In the video I walk through three options. Escape: delete it and use something else, knowing American chatbots aren't a vault either. Defend: keep it, but treat it like a postcard. Public information is fine; health, money, work documents and passwords are not. Attack: find out where else it's hiding by checking which AI model the apps you already use are built on, and where they send your data. Is DeepSeek on your phone, and after watching this, does it stay? Tell me in the comments. I read them. If this helped, give it a like, subscribe, and turn on the bell so you don't miss the next one. Share it with someone who uses free AI apps. And if you'd like to support the channel, Patreon is linked below. Keywords: DeepSeek, DeepSeek ban, DeepSeek government ban, Homeland Security DeepSeek order, is DeepSeek safe, DeepSeek privacy, DeepSeek data China, DeepSeek V4.1 Flash, free AI app, Chinese AI, AI privacy, AI news explained, FASCSA order, open source AI models, AI chatbot safety 👇 Connect with The AI Guide: Website: www.theaiguide.ai YouTube: https://www.youtube.com/channel/UCamFJyTjb_kqUbNpVUVwQeg Patreon: www.patreon.com/theaiguide Facebook: www.facebook.com/davidtheaiguide Instagram: www.instagram.com/theaiguide LinkedIn: https://www.linkedin.com/company/the-ai-guide-on-youtube/ Sources: 1. TechCrunch, US orders federal agencies to remove China's DeepSeek AI (Oct. 20, 2026): https://techcrunch.com/2026/10/20/us-orders-federal-agencies-to-remove-chinas-deepseek-ai-from-their-systems/ 2. BleepingComputer, DHS bans DeepSeek products across US federal civilian agencies: https://bleepingcomputer.com/news/security/dhs-bans-deepseek-products-across-us-federal-civilian-agencies 3. The Hacker News, US orders DeepSeek removed from federal systems: https://thehackernews.com/2026/10/us-orders-deepseek-removed-from-federal.html 4. Federal Register public inspection notice 2026-21447: https://public-inspection.federalregister.gov/2026-21447.pdf 5. Cloud Security Alliance, DeepSeek FASCSA order research note: https://labs.cloudsecurityalliance.org/research/csa-research-note-dhs-deepseek-fascsa-order-20261021-csa-sty/ 6. TechCrunch, DeepSeek reportedly raises $15B days after US restrictions (Oct. 22, 2026): https://techcrunch.com/2026/10/22/deepseek-reportedly-raises-15b-days-after-us-restrictions/ 7. Newsquawk, DeepSeek and MiniMax get license for overseas use: https://www.newsquawk.com/headlines/deepseek-and-minimax-get-license-for-overseas-use-reports-chinese-media 8. Let's Data Science, DeepSeek's new model and the DHS order: https://letsdatascience.com/blog/deepseek-new-model-saturday-dhs-banned-agencies-monday #DeepSeek #AIPrivacy #AINews #ChineseAI

93% of companies blew past their AI budgets in six months. That bill may decide who wins AI: the giant "Fat Models" or the Rebel Alliance. On June 30, 2026, Shai Goldman's Brex Startup Community newsletter ran an item called "Rebel Alliance vs. Fat Models." It summed up a post from Union Square Ventures, the New York venture firm, written by partner Nick Grossman on June 25. Goldman added a warning that turned out to be early: many companies had already gone over their 2026 token and inference budgets, and the year was only half done. Since then, McKinsey, Bloomberg, and TechCrunch have filled in the rest. So who's actually winning? First, the two camps. The "Fat Models" are the biggest AI labs, which want to own every layer, from the model to the apps and agents you use. Goldman named OpenAI, Anthropic, and Google's Gemini, plus SpaceX and its xAI unit. The "Rebel Alliance" is USV's name for hundreds of smaller companies that each build one piece of the stack: memory, routing, payments, identity, browsers, and the tools that let AI agents do work. USV argues that AI is too big for any one company to own, the same way the PC beat the mainframe and the web unbundled the desktop. Now the update. The Fat Models have only gotten fatter since June. SpaceX closed its sixty billion dollar purchase of Cursor, the popular AI coding app, on August 14, 2026, one of the biggest tech deals ever. OpenAI is in talks to raise up to thirty billion dollars at a 1.25 trillion dollar valuation, TechCrunch reported on October 22. Anthropic raised money at a 965 billion dollar valuation in May and is reportedly eyeing an IPO. By that measure, the giants look unstoppable. Here's the turn. The thing powering the Fat Models, usage, is also what's pushing customers away. On September 30, McKinsey's Sachin Chitturu said 93 percent of enterprises reported going over their AI budgets in the past six months, citing a FinOps survey. One in five organizations McKinsey surveyed had cut back AI use because of costs. His line: "The cost of tokens is not reducing as much as the usage of tokens is increasing." AI agents can use five to thirty times more tokens per task than a simple chat, per a Gartner estimate he cited. That's the bill Goldman saw coming in June. So what do the people paying the bill do? McKinsey expects most company AI work, roughly 80 to 85 percent, to move to open-weight models, the kind anyone can download and run, keeping only 10 to 15 percent on frontier models. Open-weight models cost roughly a third to a tenth as much per token. That's exactly the mix-and-match world the Rebel Alliance is built for. Goldman's view is that both sides can win, since the market is still early, and 2027 budget season is when boards and finance teams will start shopping around. The catch is that none of this is settled. The big labs still lead on the hardest tasks, they're cutting prices, and they keep buying the tools rebels build. McKinsey's split is a forecast, not a fact. And valuations near a trillion dollars say investors are still betting on the giants. What does this mean for you? If your company uses AI, the tools you get at work in 2027 may come from a mix of providers instead of one big name, picked partly on price. That could mean cheaper and more choice, or more confusing. If you run a small business, it's a good time to ask what you're paying per task and whether a smaller model would do the same job. The AI race isn't only about who builds the smartest model. It's also about who sends the smallest bill. So who's your money on: the Fat Models or the Rebel Alliance? Tell me in the comments. Like, subscribe, and hit the bell to stay ahead on all things AI. Share this with someone who'd want to see it. Support the channel on Patreon (link below). Keywords: Rebel Alliance AI, Fat Models, Union Square Ventures, Brex, AI costs, AI budgets, open-weight models, McKinsey AI, SpaceX Cursor, OpenAI valuation, Anthropic, AI agents, AI news, artificial intelligence 👇 Connect with The AI Guide: Website: www.theaiguide.ai YouTube: https://www.youtube.com/channel/UCamFJyTjb_kqUbNpVUVwQeg Patreon: www.patreon.com/theaiguide Facebook: www.facebook.com/davidtheaiguide Instagram: www.instagram.com/theaiguide LinkedIn: https://www.linkedin.com/company/the-ai-guide-on-youtube/ 🔗 Sources: [1] Brex Startup Community: https://brex.beehiiv.com/p/rebel-alliance-vs-fat-models [2] USV, The Rebel Alliance: https://www.usv.com/writing/the-rebel-alliance-2 [3] Computer Weekly (McKinsey): https://www.computerweekly.com/news/366651239/Token-bills-to-push-most-enterprise-AI-workloads-onto-open-weight-models [4] Advisor Perspectives (Bloomberg): https://www.advisorperspectives.com/articles/2026/08/14/spacex-completes-60-billion-cursor-acquisition [5] TechCrunch: https://techcrunch.com/2026/10/22/openai-working-on-new-30b-raise-at-1-25t-valuation-ahead-of-ipo/ #AINews #OpenSourceAI #AIcosts #ArtificialIntelligence #TheAIGuide