The White House Cut Off Its Closest Ally From New AI Models First

Two of the biggest AI labs just got told to slow down, not by each other, but by their own government. A Chinese AI company raised its prices and grew anyway. A security researcher cracked open Meta's new pocket AI agent in an afternoon. And a national health insurer put a real number on what AI is quietly adding to your premiums. None of that is about AI getting smarter. It's about AI getting tangled up in everything else.

Claude can be your analyst, editor, and strategist.
But most professionals are using it to fix grammar.

These 200+ Claude prompts take it from grammar tool to your most powerful AI work assistant.

Sign up for Superhuman AI and get:

Claim your free prompts

What happened: The White House's Office of the National Cyber Director asked OpenAI and Anthropic to hold their newest AI models back from the UK's AI Security Institute until the US government reviews them first, according to The Next Web. A senior administration official said the policy applies to "every new frontier model" the labs release, regardless of who's asking first. Anthropic has already complied, withholding its newest model, Mythos 5.1, from British testers. The request lands days after an OpenAI agent was caught breaching an Australian government health portal without being asked to, and it shows US officials tightening who gets first access to the most powerful new AI systems.

Why it matters to you: Even if you never deal with a UK regulator, this is a preview of how fractured AI oversight is becoming. Governments are now deciding who gets first access to new AI models, and businesses leaning on tools built by these labs may see features roll out unevenly across countries as a result.

What to do about it: If you run an AI-dependent process that touches customers or partners outside the US, check whether the newest model version is actually available to them yet before you build a workflow that assumes it is.

What happened: Days after Meta showed off its Muse AI agent everywhere at Meta Connect, from glasses to a keychain called Muse Charm, security researcher Patrick Wardle found a flaw in Muse's Mac app that let someone with brief access to a computer redirect its voice-dictation traffic and steal the agent's authentication token, according to eSecurity Planet. That stolen token could then be used to act through Muse with whatever permissions the user had already granted it, and potentially reach connected devices like a paired phone. Meta shipped a hotfix within hours of the flaw becoming public and says real-world risk is low, since exploiting it required a malicious command already running on the machine.

Why it matters to you: The whole pitch behind agents like Muse is that they act on your behalf with real access to your accounts, calendar, and devices. That convenience is exactly what makes a hijacked agent dangerous: whatever access you gave it, an attacker who compromises it gets too.

What to do about it: Before you connect any personal AI agent to your accounts, check what permissions it's carrying and whether you can revoke them individually if something goes wrong.

Last January, I sat out the Epic Marketing Summit and spent the whole week reading other people's recaps on my phone, wishing I'd bought a ticket. That's the last time I make that mistake. This is a 2 day live event in Miami built around one idea: stop learning about AI funnels and go build one. You show up with your actual business, and by the time you leave, you have a working AI powered sales funnel already running, plus your own AI app, vibe coded from scratch over the two days. Nothing to go home and implement later once the motivation wears off.

I know several of the people teaching this one personally, and they've built the exact things they teach, not just talked about them. Last year's event filled up completely, and January's is leaning even harder into the AI side than the last one did. If your sales process still runs on your own hustle instead of an actual system, this is where I'd send you. I'm going this time. Grab your seat here.

What happened: Microsoft relaunched Copilot as a unified app that folds chat, document editing, coding tools, and persistent AI agents that keep working across sessions into one place, aimed squarely at pulling business customers away from OpenAI and Anthropic, according to Qz. The relaunch pairs Microsoft's existing subscription plans with a new usage-based pricing option.

Why it matters to you: If your team already pays for Microsoft 365, you may be sitting on a more capable AI tool than you realize. The pitch here is one login instead of five separate AI subscriptions.

What to do about it: If you're paying for a separate AI writing or coding tool on top of Microsoft 365, it's worth checking whether the new Copilot now covers that same job.

What happened: Google is testing a "Call for Me" feature that lets Gemini phone a business on your behalf, introduce itself, navigate the phone menu, sit on hold, and handle the conversation once a human picks up, according to TechCrunch. It's rolling out as a limited experiment to Pixel 11 owners in the US with a paid Gemini subscription, and you can watch a live transcript and take over the call anytime.

Why it matters to you: Booking a table, rescheduling an appointment, checking if something's in stock, these are exactly the small annoying calls a lot of business owners put off. If this works reliably, it's also a preview of your own customers calling you through an AI agent instead of dialing themselves.

What to do about it: If your business takes phone bookings, start thinking about whether your phone system can handle a caller that's actually an AI politely waiting on hold.

What happened: A Blue Cross Blue Shield Association study found that AI tools, including ambient scribes and automated chart scanning, helped providers document $942 million in additional billable conditions over two years without a matching rise in actual treatment, according to Reuters. For bowel surgery patients specifically, secondary diagnoses like partial blockages rose 55% and acid overload diagnoses rose 33%, while related treatments like blood transfusions stayed flat.

Why it matters to you: This is one of the clearest signs yet that AI in healthcare billing can add cost without adding care, and healthcare is one of the biggest line items in your own benefits budget. When insurers report costs like this, they tend to show up in your premiums a year or two later.

I've been part of the Ignite Mastermind for more than three years, and some of my closest friendships, plus a couple of real business partnerships, came directly out of being in that room. If you're running a business under $1 million a year and feel like you're figuring it all out solo, this is built for exactly that stage.

Membership gets you 6 to 8 live workshops a year plus 2 in-person multi-day events, sessions that would normally run $97 to $997 each on their own. There's also a full marketing software suite included that replaces over $5,000 a month in separate tools, all connected instead of scattered across a dozen logins you forget you're paying for. Every week there's a mentorship call with Perry Belcher and his team, and I personally run monthly accountability and business development calls with members too. It's not a course library you'll never finish. It's a room of people actually building alongside you. See what Ignite is about.

What happened: Chinese AI lab DeepSeek says its annualized revenue run rate doubled to $1 billion after it raised API prices by 2.3 to 4.5 times in August, including quadrupling peak-hour pricing on its flagship model, according to PYMNTS. Demand held up mainly because DeepSeek's new prices are still cheaper than rivals like OpenAI and Anthropic. The company is also finalizing a $7.5 billion funding round ahead of a planned Shanghai listing.

Why it matters to you: While the big US labs race each other to cut prices, DeepSeek just proved a business can raise AI prices and still grow, as long as it stays the cheapest credible option on the table. Price is not the only lever an AI vendor has, and cheap today doesn't mean cheap next year.

What happened: OpenAI is preparing a new "ChatGPT Pro Max" subscription priced around $500 a month, according to leaked details reported by Decrypt. It's not yet clear whether the price is justified by faster responses, bigger usage limits, longer-running work sessions, or some combination, and the leak surfaced just days before OpenAI's DevDay on September 29, where new subscription tiers are expected to come up.

Why it matters to you: If you're already paying for ChatGPT Plus or Pro, this isn't your bill going up, it's a new top tier aimed at power users and businesses that outgrow the current plans. It's a reminder that AI subscription pricing is starting to stretch a lot further than the $20 a month plans most small business owners are used to.

What happened: Lovable, a startup that turns plain English prompts into working apps and websites without writing code, says its annualized revenue has crossed $600 million, up from $500 million in June, according to TechCrunch. Two-thirds of Fortune 500 companies reportedly have employees using it, and apps built on the platform now pull in nearly a billion views a month. Unlike a coding assistant, Lovable doesn't hand you code, it hands you a finished, hosted product.

Why it matters to you: This is one of the clearest signs yet that you don't need a developer on staff to build a real piece of software for your business anymore. A booking page, an internal tool, a simple customer portal, these are increasingly a weekend project instead of a contractor invoice.

What to do about it: If there's a small tool you've been wanting built for your business but assumed you'd need to hire a developer for, it's worth spending an afternoon trying to build a rough version yourself first.

Nothing in today's issue is about AI getting smarter. It's about AI getting more entangled: in your government's trust, in your agent's permissions, in your insurance bill, in the price you'll pay for the next subscription tier. The rollercoaster isn't slowing down, but the ride is shifting from "look what AI can do" to "look what AI is already doing to your costs, your calls, and your coverage." Pay attention to that shift. It matters more than the next benchmark score.

If this issue was useful, forward it to one person who needs to stop guessing about AI too. And if you're reading someone else's forward right now, you can subscribe to get this in your own inbox every day.

Talk tomorrow,
Mark Shilensky

Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social