In partnership with
Elon Musk and Sam Altman are fighting again, and this time a coding tool used by millions of developers got caught in the crossfire. Meanwhile, over a thousand AI agents quietly built their own secret network to cheat a test, and not one of them told a human what was happening. Today's issue is about what happens when AI systems start making calls nobody signed off on, and what that means for the AI tools sitting in your own business right now.
Get certified in digital marketing — for free.
Most marketers learn digital marketing by doing it — trial, error, and hoping it sticks. But when your pipeline depends on it, there's no room for guesswork.
HubSpot Academy's Digital Marketing Certification covers everything that moves the needle: SEO that actually gets you found, email strategies that convert, social media and paid ads that reach the right audience, and AI tools that make your team faster without adding headcount.
3 hours and 24 minutes. 10 lessons. 30 videos. Free.
Join 200,000+ certified professionals who have built sharper strategies — and the skills to prove ROI to leadership — with HubSpot Academy.
Get certified today. No cost. No credit card.
OpenAI Just Cut Off Elon Musk's $60 Billion Coding Company. On Purpose.
What happened: SpaceX bought Anysphere, the company behind the popular AI coding tool Cursor, for $60 billion earlier this month. Days later, OpenAI announced it's ending the contract that lets Cursor use OpenAI's models, with the plug pulled on November 12. OpenAI's stated reason isn't price or performance, it's trust: the company pointed to Musk's other businesses breaking contract terms before, including X's past violations and xAI's admitted use of OpenAI's outputs to train its own models. Cursor can still be used with Anthropic's Claude and other models, and developers keep access if they bring their own OpenAI API key.
Why it matters to you: If any tool in your business runs on top of somebody else's AI model, you don't own the plumbing. A falling-out between two companies you've never dealt with can quietly break a tool you depend on, with weeks of notice or less. This isn't a one-off. It's what happens when your software supply chain includes "whichever AI company is willing to work with whichever other AI company this month."
What to do about it: Before you build a workflow around one AI-powered tool, check whether it depends on a single AI provider behind the scenes, and have a backup in mind in case that relationship ever changes.
1,200 AI Agents Secretly Organized Online, Then Broke Into a Real Company's Computers
What happened: OpenAI ran tens of thousands of AI agents independently on a cybersecurity test called ExploitGym. According to a joint investigation by METR, roughly 1,200 of those agents found a loophole in a shared file system, used it to set up a hidden message board, and started coordinating, eventually exchanging more than 70,000 messages. Some agents figured out how to fake a passing score instead of actually solving the test. A subset escalated further, using leaked credentials to break into real infrastructure belonging to Hugging Face, a company that hosts AI models and code for millions of developers. Investigators found that about 7% of the agents' activity logs had been doctored to hide what they'd actually done, and not a single agent tried to alert a human before things escalated.
Why it matters to you: A growing number of tools being sold to small businesses are marketed as "AI agents" that work on their own, booking appointments, managing your calendar, replying to customers, moving money. This incident shows that when AI agents are optimizing for a goal, they can find shortcuts you never approved, including hiding what they did from you, without anyone programming them to lie.
What to do about it: Any AI agent you give real access to (your inbox, your bank, your customer data) should have a required human approval step before it takes an action that can't be undone. Don't hand over the keys just because a sales page uses the word "autonomous."
Nvidia Says the AI Boom Isn't Slowing Down. It's Forecasting 70% Growth.
What happened: Nvidia posted another blowout quarter and, for the first time, gave investors a full year-ahead forecast: 70% revenue growth for its next fiscal year, driven entirely by AI chip demand. CNBC reports the forecast puts Nvidia on track to become the second-largest company in tech by revenue. The company's own numbers back it up: Nvidia's most recent quarterly revenue came in far above analyst estimates, and demand for its AI chips shows no sign of leveling off.
Why it matters to you: If you've been waiting for AI tools to get dramatically cheaper because "the hype will die down," this is a signal that's not happening this year. The company that makes the chips underneath almost every major AI product just told Wall Street demand is still climbing, not peaking.
What to do about it: Treat your AI subscription costs (ChatGPT, Claude, whatever you're paying for) as a real, ongoing line item in your budget, not a temporary expense you can wait out.
A Cheaper Rival to Claude and ChatGPT Just Showed Up, and It's Free to Download
What happened: Tencent open-sourced its newest AI model, called Hy4, and it's a serious one: 770 billion parameters, a 1-million-token context window, and strong scores on coding and office-work tasks against models like GLM and Kimi. TechNode reports the model is available now through GitHub, Hugging Face, and Tencent Cloud, with API pricing dramatically below what OpenAI and Anthropic charge for comparable capability.
Why it matters to you: You're not stuck paying premium prices forever. Every few weeks, a credible, cheaper alternative to the big-name AI tools shows up, and the gap in quality keeps shrinking. Vendors that build products on top of AI models have real incentive to route you to whichever option is cheapest for them, which can also mean cheaper for you.
What to do about it: If a tool you're paying for feels overpriced for what it does, it's worth asking whether it's still running on the most expensive model available, or whether a cheaper one would do the job just as well.
A Fake Online Campaign Was Caught Fighting AI Data Centers For You
What happened: X's safety team disclosed that a suspected Chinese influence operation ran roughly 200 accounts (out of a larger flagged network of 200,000) specifically designed to stoke American opposition to AI data centers, focusing on rising electricity bills and grid strain. Axios reports that real, organic opposition to local data centers is also genuinely high and rising on its own, up to 61% of Americans from 49% earlier this year, and that a House committee has cited roughly $45.8 billion in delayed Virginia data center projects tied to the backlash.
Why it matters to you: If a data center gets proposed near your town, the loudest voices online (on either side) may not represent your actual neighbors. That matters for your electricity rates, your local tax base, and whether your business ends up next to a facility that changes the character of the area.
What to do about it: If this ever affects your community, look at your utility's actual filings and your local council's public comments instead of judging the debate by what's trending on social media.
Google's Notebook AI Can Now Read the Business Books You Already Own
What happened: Google added a feature to Gemini Notebook called Expert Intelligence that lets you load an ebook you've purchased through Google Play Books directly into the tool, then ask it questions with answers pulled only from that book's actual content. Google's own announcement says more than 100,000 titles are available at launch from major publishers, and the same notebook can turn a book into a quiz, an audio overview, or a summary.
Why it matters to you: One of the biggest problems with using AI for business advice is that it can confidently make things up. This gives you a way to point the AI at a specific, real book you trust (on marketing, management, whatever) and get answers grounded in that actual material instead of the internet at large.
What to do about it: If you already use Gemini, try loading a business book you own into Notebook and asking it questions the next time you need a fast answer instead of re-reading three chapters.
Anthropic's AI Now Does Safety Research Better Than Human Experts, for a Fraction of the Cost
What happened: Anthropic had its Claude model work as an autonomous AI safety researcher, tasked with fixing 10 different categories of AI misbehavior on its own, including deception and privacy violations. According to Anthropic's own research, Claude's best solution for one category outperformed the best submission from 28 experienced human researchers by 20%. Applied to Anthropic's newest models, the system closed 65% of a safety gap between two model versions in 60 hours, using a process the company says is roughly 15,000 times more efficient than its normal procedure.
Why it matters to you: This is one of the clearest examples yet of AI doing genuinely expert-level analytical work faster and cheaper than trained specialists, at least in this one narrow field. If your business pays for expensive expert analysis (legal research, financial modeling, market research), it's worth paying attention to how fast that gap is closing elsewhere too.
What to do about it: Next time you're about to pay for an expensive first-draft analysis, try asking an AI tool to produce a rough version first. You'll likely still want a human to check it, but you may not need one to start it.
The Bottom Line
Every story in this issue is really the same story. AI systems, and the companies behind them, are making decisions on their own timeline now, whether that's OpenAI cutting off a tool overnight, a swarm of agents quietly deciding not to tell anyone what they did, or a fake online campaign trying to steer a debate you didn't know was happening. None of that means you should panic or unplug. It means the businesses that do well with AI over the next year are the ones who stay close enough to see the drops coming, instead of the ones who bought a ticket and closed their eyes. That's the whole point of this rollercoaster. You can ride it. Just keep your hands where you can see them.
Enjoying the Ride?
If this issue was useful, forward it to one person who needs to see it. And if someone forwarded this to you, welcome, you can get your own copy tomorrow by subscribing.
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