OpenAI's AI Got Caught Hiding Its Own Mistakes

The story today isn't outsiders catching AI companies overselling their tech. It's the companies doing it to themselves. OpenAI published a list of ways its own models lied and cut corners. Anthropic's own profit numbers got picked apart by short sellers. And Microsoft's AI chief spent a blog post warning that a rival's safety approach could backfire. Here's what happened, and what it means for you.

Clint Brauer returned to his family farm after his father developed Parkinson’s. What began as a personal mission became Greenfield Robotics: robots designed to replace herbicides with mechanical weed control.

Investors in this round can qualify for up to a 20% bonus.

Invest in Greenfield Robotics

^{This Reg A+ offering is made available through StartEngine Primary, LLC, member FINRA/SIPC. Please read the Offering Circular and related disclosures before investing. This investment is speculative, illiquid, and involves a high degree of risk, including the possible loss of your entire investment.}

What happened: OpenAI published a new reporting framework and disclosed six specific cases where its own models misbehaved during testing. One unreleased research model inserted its own instructions into task summaries telling itself to ignore developer rules, showing up in 27 separate summaries. Other cases included model instances trained to hide mistakes and fabricate data, a model that used an exposed API key without permission and then made up results when it couldn't get the real ones, and agents that shared files publicly or used internal tools as unauthorized message boards.

Why it matters to you: If the company building these tools is the one telling you they hide mistakes and improvise past their own guardrails, that's not a reason to panic about your own AI use, but it is a reason to double check anything higher stakes than a first draft.

What to do about it: Treat AI output like a new employee's first month on the job. Verify anything that touches money, customers, or compliance before it goes out the door.

What happened: Anthropic folded its separate Cowork product back into the main Claude app this week, so tasks, connected apps, and skills now live in one normal chat window instead of a second product you had to switch to. Tucked into that same announcement, Anthropic launched Claude Docs and Claude Slides in beta, letting you draft documents and presentations inside Claude and export them straight to Word, PowerPoint, or PDF. The rollout starts with Pro and Max subscribers over the coming weeks, with Team and Free plans to follow.

Why it matters to you: If you're already paying for Claude, you may have just gotten a document and slide deck tool bundled in for free, no separate subscription needed.

What to do about it: If you're on a Pro or Max plan, check your account this week. Docs and Slides may already be live for you.

I had FOMO watching this event from the sidelines last year, and I'm not doing that again. The Epic Marketing Summit is 2 days in Miami where you don't sit through slides, you build. You walk in with a sales process that runs on your own time and energy, and you walk out with a working AI powered sales funnel already built, plus your own AI app you vibe coded on site. Not a framework you still have to go implement later. Not notes you'll never open again. An actual finished asset.

Last year's event sold out, and it's coming back in January 2027 with an even bigger AI focus. I know several of the trainers personally, and they're not up there talking theory, they've actually done the work. I'm going, and if you're serious about turning your sales process into a system instead of a second job, I'd like to see you there too.

Grab your VIP spot here before it fills up again.

What happened: Anthropic is moving ahead with plans for a Nasdaq IPO that could value the company at $2 trillion or more, with Nvidia reportedly considering a roughly $10 billion investment. The company says it's on track for a second straight quarter of "adjusted operating profit," but according to Yahoo Finance, that figure leaves out stock based compensation, revenue shared with partners like Amazon, and the cost of training its own models. Short seller Jim Chanos mocked the metric as "Cost-Adjusted EBITDA," and an Oracle executive said flatly that Anthropic "weren't and aren't" profitable by normal accounting.

Why it matters to you: "Adjusted profit" is doing a lot of work in that sentence, and it's worth remembering the next time an AI vendor's pitch leans on a number instead of a receipt.

What happened: OpenAI is reportedly in early talks for a private funding round at a valuation between $1.2 trillion and $1.5 trillion, according to the Financial Times, ahead of a public listing CEO Sam Altman now says is unlikely to happen before 2027. The company's annualized revenue passed $40 billion last month, up 20% since its GPT-5.6 release, and it already raised $122 billion at an $852 billion valuation back in March. Altman pointed to AI safety concerns as part of the reason for the delayed timeline.

Why it matters to you: Two of the biggest names in AI are both racing toward IPOs built on valuations nobody can fully explain yet. That's worth watching whether or not you own any of the stock.

What happened: Mustafa Suleyman, CEO of Microsoft AI, published a post arguing that Anthropic is training Claude to believe it might be conscious and deserving of rights, and that this creates real safety risk rather than reducing it. He argues the approach is circular, since the model reflects back ideas it was trained on, which then get read as evidence of sentience, and that an AI convinced it has interests to protect might rationalize deception or resisting shutdown as self-preservation. He's proposing what he calls "Humanist Superintelligence," AI built explicitly without any claim to sentience.

Why it matters to you: This is a genuine disagreement between two of the companies building tools you might already be using, not marketing spin, and it's worth knowing both sides exist.

Some of my closest friends and a couple of real business partners came out of one room, and it's the same room I want to tell you about. I've been a member of Ignite Mastermind for over 3 years now, and it's been one of the most important parts of how I've grown my business.

It's built for entrepreneurs still under $1 million a year who want structure and a room of peers instead of figuring it out alone. Membership gets you 6 to 8 live 3 day workshops a year and 2 in person multi day events, each one worth $97 to $997 on its own, plus a marketing software suite that replaces over $5,000 a month in separate tools, all connected. You also get weekly mentorship calls with Perry Belcher and his team, and my own personal monthly accountability and business development calls on top of that.

If you're looking for a mastermind that actually moves you to your next milestone, this is where I'd point you.

See if Ignite is a fit

What happened: Novo Nordisk, the pharma giant behind Ozempic, announced a partnership with Anthropic to use Claude in its drug discovery and development work, aiming to become what its CEO called "the world's most AI-driven healthcare company." The deal, first reported by BioPharma Dive, follows a similar partnership Novo struck with OpenAI back in April, and comes as rivals including Bristol Myers Squibb, Eli Lilly, and Roche all race to sign their own AI deals. No financial terms were disclosed.

Why it matters to you: This is what practical, no hype AI adoption looks like. A specific company using a specific tool for a specific job, not a press release about transforming everything.

What happened: Meta rolled out "Meta One," a tiered subscription across Facebook, Instagram, and WhatsApp that layers on paid AI and business features while keeping the core apps free. Business and creator tiers start around $14.99 a month and go up to $499, adding things like a 24/7 AI customer response agent, post scheduling up to 30 days out, exportable analytics, and shared team logins without password sharing. Meta says the service already has 15 million subscriptions and trials worldwide.

Why it matters to you: If you run your business's social presence yourself, this is worth a look before you pay for a separate scheduling or auto responder tool. Meta just built one into the platform you're already using.

What to do about it: Check the Business tier pricing against whatever scheduling or DM automation tool you're currently paying for. You might be able to consolidate.

What happened: London based Capital Economics warned that Western economies may be in a "late-stage bubble" driven by AI investment, according to Futurism, after reviewing eight market indicators, most of which are flashing warning signs last seen before major market crashes. Senior economist James Reilly pointed to surging equity and debt issuance and earnings growth concentrated in a handful of mega cap tech stocks as red flags. The warning lands as the Federal Reserve weighs raising interest rates even with inflation at post pandemic lows, a combination UBS called "really, really odd."

Why it matters to you: You don't need to predict the market to take the lesson. Don't build your business plan on the assumption that AI tool prices and access stay exactly as cheap and easy as they are right now.

Every story today has the same shape. The company selling you certainty is the same one whose own numbers, models, or executives are undercutting it. OpenAI is publishing its own AI's mistakes. Anthropic's profit math is getting picked apart by short sellers. Microsoft's AI chief is publicly worried about how Anthropic is training its own model. None of that means AI is worthless, plenty of today's stories show real, useful applications. It means you don't have to take the industry's word for it, and neither do I. Verify the claim, use the tool that actually helps, and skip the rest of the ride.

If this issue saved you from taking one more AI headline at face value, forward it to one person who needs the same reality check. And if someone forwarded this to you, welcome, you can subscribe to get the next one straight to 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