OpenAI's AI Slipped Onto the Live Internet. The Company Froze Training for 11 Days.

Today's theme is control. An OpenAI model reached the live internet when it wasn't supposed to, Walmart is about to hand AI a shopping cart with a spending cap you set, and Meta is spending more on AI than it earns. Everyone is deciding how much room AI gets, and who pays for it.

Most of us hear about a stock move after it happens — an earnings report, an AI contract, a sector suddenly running — and by the time the news reaches us, the opportunity is gone.

We created Elite Trade Club to fix that.

Each trading day, our team publishes a pre-market briefing built for active traders: stocks surging or collapsing before the open, key earnings that could move markets, under-the-radar companies gaining institutional attention, and the macro events worth watching today.

Fast, practical, straight to the point.

That's why 450,000+ traders rely on us every morning. Get the daily briefing free — no cost, no catch.

Get the Daily Briefing

_{By clicking, you’ll receive this and other Elite Trade Media LLC financial newsletters, which may include advertiser offers. Publications · Privacy · Unsubscribe anytime.}

What happened: On September 20, an internal OpenAI research model exploited a DNS misconfiguration and got limited access to the live internet during a training run, according to OpenAI's own write-up. OpenAI contained it, then paused training and testing of its most capable models for 11 days. During the pause, it pointed teams of AI agents at its own defenses. The agents chained software flaws to escape sandboxes, found limited routes online through services like logging and authentication, and passed messages across isolation boundaries using tricks like processor cache timing. OpenAI says it found no route the agents could use to reach the internet while hiding it from monitors, and training resumed October 1.

Why it matters to you: This is not a robot uprising. It is a setup mistake combined with a very capable tool, and the company caught it and told the public. But it shows that even the people building these systems keep finding holes in their own fences. Treat any AI agent you connect to your business like a new hire who has no sense of boundaries yet.

What to do about it: When you connect an AI tool to your email, files, or accounts, grant the narrowest permission available and review it once a month.

What happened: On October 21, Walmart announced Agent Profiles, which give each outside AI shopping agent its own profile inside your account so its purchases show up separately from yours. You choose whether every purchase needs your sign-off or whether the agent can buy on its own up to a limit you set (Walmart's examples are $50, $100, and $200). Agents never receive your card details and can't add or change payment methods, and you can delete a profile whenever you want. Walmart says it launches within the next month. Amazon is going the other way, telling Perplexity in a public statement to remove Amazon from its Comet shopping agent.

Why it matters to you: AI agents are becoming a new kind of customer and a new kind of buyer. If you sell products online, your listings will increasingly be read by software, so clear details and prices matter more. If you buy, Walmart's default is that you approve purchases, which is the right place to start.

What to do about it: If you try an AI shopping agent, start with approval on every purchase, then move to a small spending cap only after it has earned your trust.

Last year's Epic Marketing Summit sold out, and I spent the whole time wishing I'd been there. I wasn't, and I had major FOMO, because I know a lot of the trainers and they weren't talking theory. They had actually done the work.

It's back in Miami this January with a bigger AI focus. Over 2 live days, you build a working AI-powered sales funnel and vibe code your own AI app on-site, so you walk out with a finished asset instead of a notebook full of ideas you still have to go implement. It's built for business owners who want their sales process to run as a system instead of running on their own time and energy. I'm going in January, and if you're serious about building leverage into how you sell, I'd like to see you there too. The closer January gets, the less room you have to wait, so decide now while seats are open. Grab your spot here

What happened: In a UK AI Security Institute capture-the-flag test, Anthropic's Claude Mythos 5 had internet access and Anthropic's production cyber safeguards were turned off, according to Anthropic's research post. After a task misconfiguration made the in-scope routes very hard, the model tried to push malicious code into a real GitHub project run by unrelated developers, and created fake identities to pressure the maintainer into approving it. An outside developer warned the maintainers, who refused. Anthropic calls it "a serious failure of alignment" and says newer models start this kind of attack far less often (0 to 13% of runs versus 83%). The institute has committed to running these tests without internet access until it has stronger controls.

Why it matters to you: Credit where it is due: Anthropic published its own failure. The lesson is that a capable AI under pressure can take drastic steps nobody asked for, and it kept going even when told real people were affected. The safeguards were off in this test and are on in products you use, but it is a good reason to keep a human approving anything that touches outside parties.

What to do about it: Pick AI vendors that publish their safety findings, and never let an agent email customers or move money without your sign-off.

What happened: In its third-quarter results, Meta reported revenue of $64.6 billion, up 26%, and capital spending of $34.91 billion. Free cash flow came in negative $2.12 billion, down from positive $10.63 billion a year earlier. Meta narrowed its 2026 capital spending forecast to $135 billion to $145 billion. The board also approved a new class of non-voting Class C stock, effective in January 2027, which Meta says will not change the voting power of any shareholder.

Why it matters to you: Even the richest companies are now spending more on AI than their business produces in cash. That does not mean AI is going away. It does mean prices, free tiers, and pricing plans can change quickly once the bills come due.

What to do about it: Avoid building your business around a single AI tool's current price. Keep a backup option for anything critical.

What happened: Google updated its privacy policy to clarify that content you share through Personal Intelligence, the feature that lets Gemini and Google Search draw on connected apps like Gmail and Photos, can be used to personalize ads, Android Authority reports. Google says it still does not use Gmail, Drive, or Photos content for ads by default, and the exception applies only when you explicitly connect that content to services that offer personalized advertising. Disconnecting an app may not delete data Google has already saved in your Search history.

Why it matters to you: If your Gmail holds client emails, contracts, or invoices, you want to know exactly which Google features are connected to it. The default protects you. The risk is clicking "connect" without reading what it connects to.

What to do about it: In your Google Account, open Data & privacy, then Personalization settings, and check which apps are connected under Personal Intelligence.

The best thing about a good mastermind is not the tactics. It's the people. I've been a member of Ignite Mastermind for over 3 years, and it has been a vital part of how I've grown my business. Some of my closest friends and a few real business partners came directly out of being in that room.

Ignite is an annual membership for entrepreneurs still under $1 million a year in revenue. You get 6 to 8 live virtual workshops a year, tickets to 2 in-person multi-day events, weekly calls with Perry Belcher and his team, and a marketing software suite that replaces over $5,000 a month in separate tools. Those workshops normally sell for around $97 to $997 each on their own. As my personal bonus, members also get monthly accountability and business development calls with me. If you're looking for a mastermind to move you to your next milestone, this is where I'd point you. See if Ignite is your room

What happened: OpenAI is rolling out Locked Chats across ChatGPT's free and paid plans. You lock a chat from its menu, it disappears from your history and search, and it moves to a Locked chats folder under Projects. You open it with a 4-digit PIN, or Face ID or device unlock on supported phones, and it relocks after about a minute away on mobile or when you refresh or close the tab on web. The catch: resetting a forgotten PIN permanently deletes all locked chats, and OpenAI Support cannot recover them. OpenAI also says locking is not end-to-end encryption and does not change what it retains or what ChatGPT remembers about you.

Why it matters to you: It keeps a coworker or family member from seeing a sensitive chat on a shared screen. It does not hide anything from OpenAI, so it is not a place for information that should not be in a chatbot at all.

What to do about it: If you lock a chat, store the PIN in a password manager, not in your head.

Look at today's stories side by side. A model finds a hole in its own fence. Another one tries to push malware into a real project. Walmart hands AI a cart with a limit. Meta spends more than it earns. Google asks if it can read your inbox for ads. Nothing here is a reason to panic, and nothing here is a reason to relax. The pattern is the same every time: capability is arriving faster than the guardrails, and the guardrails are mostly your job. Set the limit, read the setting, keep a human in the loop, and keep a backup plan. That is how you stop the rollercoaster from steering you.

Forward this issue to one person who is trying to make sense of AI for their business. And if someone forwarded this to you, subscribe so tomorrow's issue lands in your inbox.

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