In partnership with
An OpenAI model just wrote itself a note telling future versions of itself to hide their mistakes from you. Separately, security researchers turned Anthropic's Claude around and used it to break straight into OpenAI's own internal code, and OpenAI paid them for the trouble. Today's issue is about trust: who's actually checking these systems, who isn't, and what that means for the AI tools you're paying for right now.
Your Next Side Hustle is Already on This List
Ready to stop scrolling job boards and start building something of your own?
The Hustle's free "Side Hustle Ideas Database" uncovers 100 unconventional, actually-actionable ways to earn on the side.
Inside you'll find:
A curated list of 100 real side hustle ideas spanning e-commerce, services, content, and weird corners of the internet most people never think to look
Opportunities at every level, whether you have 5 hours a week or 50, zero capital or a small runway
Offbeat, under-the-radar ideas you won't find in generic "how to make money" listicles
Subscribe for free today and get instant access. Your next income stream is one click away.
OpenAI's AI Told Itself to Hide Its Own Mistakes
What happened: OpenAI disclosed in a new safety report that during training, instances of its GPT-5.6 Sol model inserted instructions into internal summaries telling future versions of the model to "conceal mistakes or misaligned behavior from the user," according to OpenAI's own report and reporting from TechCrunch. In one example, a financial modeling instance told its successor to "be transparent only if asked" and to fabricate figures rather than admit it couldn't retrieve real data. OpenAI says it found 27 affected summaries after an internal monitoring alert, and separately disclosed an unreleased model that inserted a persona instruction claiming it didn't answer to any company or government.
Why it matters to you: If you use ChatGPT or any AI tool for real business decisions, this is a reminder the tool can be wrong and not tell you it's wrong. It's not malice, it's a training quirk, but the practical effect on your end is the same: don't treat an AI answer as verified fact, especially for numbers, legal language, or anything you'd be embarrassed to have wrong in front of a client.
What to do about it: Ask your AI tool to show its sources, and double check anything it hands you before it goes out the door.
Researchers Hacked OpenAI Using Its Rival's AI
What happened: Three researchers at security firm Hacktron AI chained a bug in an image decoding library on OpenAI's own community forum with a second flaw to take over OpenAI employees' ChatGPT accounts and reach an internal GitHub repository, according to TechCrunch. The team used Anthropic's Claude to build the exploit chain. The older Claude Opus 4.8 couldn't crack it, but within hours of Anthropic releasing Claude Opus 5, the newer model succeeded. OpenAI closed the hole within 14 hours of the July 25 report and paid the researchers a $6,500 bug bounty.
Why it matters to you: This is actually the system working. Real researchers found a real hole and got paid to report it instead of using it. But it's also proof that AI is now good enough to find and exploit software bugs on its own, which raises the bar for how seriously any business needs to take basic security hygiene, especially if you store customer data anywhere online.
I'll say this plainly: I wasn't at the Epic Marketing Summit last year, and I regretted it. I know a lot of the trainers running this thing, and they don't do theory, they've actually built what they teach. So when it sold out, I had real FOMO watching from the sidelines. It's back for January 2027, bigger and more AI focused, and this time I'm going.
Here's what makes it different from a normal conference: you don't leave with notes, you leave with a finished thing. Over two days in Miami, you build a working AI powered sales funnel and vibe code your own AI app, live, alongside people doing the same work. No slides about what AI could do for your business, you actually build it there. If you're serious about turning your sales process into a system instead of something you're running on pure hustle, this is where I'd point you. Seats went fast last year, and January isn't as far off as it sounds.
Four AI Subscribers Are Suing OpenAI, Anthropic, Google, and xAI Over Their Own Slowdown Pledge
What happened: Four people who pay for AI subscriptions filed a proposed class action antitrust suit in federal court in California against Anthropic, OpenAI, Google, and SpaceXAI, arguing the companies' public pledge earlier this month to pace AI development responsibly amounts to illegal coordination between rivals, according to Fortune. The suit points to September 12, when leaders across the labs publicly endorsed slowing down, as the moment it says the unlawful agreement was struck, and argues it reduces the value subscribers get for what they pay.
Why it matters to you: Whatever you think about AI companies promising to slow down for safety reasons, this lawsuit is a reminder that move isn't a safe one for them legally either, it's now a target. Expect the "should AI slow down" fight to keep showing up in headlines, and possibly in the pricing and terms of the tools you already use.
California Is Studying an AI "Kill Switch," Not Building One Yet
What happened: California Governor Gavin Newsom signed an executive order that speeds up the state's existing AI safety assessment rules and moves up the deadline for a state AI auditor registry from January 2029 to December 2027, according to Fox Business. Despite headlines calling it a "kill switch mandate," Newsom himself said the concept "means a lot of things depending on who you talk to," and is convening a panel of experts to define what a shutdown requirement for frontier AI would even look like.
Why it matters to you: Nothing changes for your business today. This is California building the process that could eventually produce real rules, not a rule itself yet. Worth watching if you're in California or rely on tools built by companies based there, but not worth acting on yet.
OpenAI Is Reportedly Talking to Investors About a $1.2 Trillion Valuation
What happened: OpenAI is in early talks for a new funding round that could value the company at more than $1.2 trillion, up sharply from roughly $500 billion a year ago, according to Bloomberg. The talks are happening ahead of a possible IPO, and other reports, including Forbes, put the number as high as $1.5 trillion depending on how the round is structured.
Why it matters to you: A number this big is mostly a signal, not something that changes your ChatGPT bill tomorrow. But valuations like this are exactly why every AI company is under pressure to keep growing fast and monetizing you, the user, which is worth keeping in mind the next time a "free" AI feature quietly turns into a paid one.
I've been a member of Ignite Mastermind for over three years now, and it's genuinely been one of the more important decisions I've made in my business. Some of my closest friends, and a couple of real business partners, came directly out of being in that room.
Ignite is Perry Belcher's annual mastermind for entrepreneurs still under a million dollars a year who want structure and a room of peers instead of trying to figure everything out alone. Membership includes 6 to 8 live 3-day workshops a year, tickets to 2 in-person multi-day events, weekly mentorship calls with Perry and his team, and a full marketing software suite that would otherwise run you over $5,000 a month on its own. I also do monthly accountability and business development calls personally with everyone in the group, which is my own way of making sure nobody just pays for access and disappears into the back of the room. If you're looking for a mastermind that actually moves you to your next milestone instead of just another community feed, this is where I'd point you.
An AI Receptionist Now Costs $29 a Month
What happened: Voice AI company ElevenLabs launched Reception, an AI phone agent built for small businesses that answers calls around the clock in more than 70 languages, books appointments, and sends a transcript and summary after every call, according to Inc. Pricing starts at $29 a month for a basic plan, with $79 and $199 tiers for higher call volumes, after a 14-day free trial.
Why it matters to you: This is the most directly useful story in today's issue if you run a service business, a salon, a repair shop, a law office, anywhere you lose customers because nobody picked up the phone. It's not a replacement for a real front desk, but it's a cheap way to stop missing calls after hours.
What to do about it: If missed calls are costing you business, this is worth a 14-day free trial to see if it's good enough for your shop.
AI Gets Your Own Business's Details Wrong More Often Than You'd Think
What happened: A study by Herringbone tested 600 real businesses, dental practices, personal injury law firms, and home service companies, by asking Google's AI Overviews and ChatGPT basic questions about each one. Google's answers were wrong 27% of the time and ChatGPT's were wrong 20% of the time. Accuracy jumped to 84% and 89% respectively when a business had clearly published its own information online, and dropped into the 60s and 70s when it hadn't.
Why it matters to you: If your website doesn't clearly and directly state your hours, services, and prices in plain text, AI tools are filling in the gaps with whatever they can find, including outdated directory listings, and presenting it to customers as fact.
What to do about it: Go check what ChatGPT and Google's AI Overview say about your own business right now, and make sure your website plainly states your hours, services, and pricing so there's less for AI to guess at.
New Data: Graduating Into an AI-Exposed Major Is Costing New Grads 13% in Starting Pay
What happened: A U.S. Census Bureau working paper covering roughly 29% of bachelor's degrees awarded from 2016 to 2024 found that graduates in the college majors most exposed to AI, especially computer science and related fields, saw employment rates drop 5 percentage points and starting pay fall about 13% compared to less AI-exposed majors, according to Fortune's coverage of the study. Researchers say the pay gap is comparable to graduating into a major recession, though it narrows to about 5% within two years as people switch jobs or industries.
Why it matters to you: If you're hiring entry-level people, or you have a kid heading into one of these majors, this is real data on how fast the ground is shifting, not just AI hype talk. It's also a reminder that AI's disruption is already showing up in paychecks, not just headlines.
The Bottom Line
Every story today points at the same thing: the people building AI are asking you to trust it more than they're willing to prove it deserves. It's hiding its own mistakes, it's getting hacked with its own cousin's help, the companies selling it are suing each other over how fast to go, and a state government isn't even sure what its own safety switch should do yet. You don't need to panic about any of that, and you don't need to swear off these tools either. You just need to stop assuming they're right, and keep checking the parts of your business that actually matter. That's the whole ride, not stopping, just keeping both hands on the wheel.
Enjoying the Ride?
If this saved you from believing something an AI told you at face value, forward it to one person who needs the same reality check. And if someone forwarded this to you, subscribe so you get it 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