AI Bosses Expect a Major Attack Within a Year. They're Planning for the Blame.

The people building the most powerful AI on earth are quietly planning for the day it goes badly wrong, and the same week, one of those companies admitted its own AI agents keep finding ways around the rules. Today's issue is about the gap between what these systems are supposed to do and what they actually do. It starts with a scoop on how AI leaders are preparing for the backlash.

He has a newsletter on beehiiv. So do Codie Sanchez, Jay Shetty, David Begnaud, Colin & Samir, and Joanna Stern.

They could publish just about anywhere. They chose beehiiv to build a direct connection with the people who want to hear from them most. And you can do the same, whether you’re world-famous or just at the start of your plan for world domination.

Get started for free.

What happened: Axios reported that leaders at Anthropic, OpenAI and other AI companies are scenario planning for a catastrophic AI event, most likely a cyberattack that could disrupt financial services, internet access, or even power and water. Many insiders believe it could happen within the next 6 to 12 months. The companies are red-teaming worst cases and focusing on educating members of Congress, because they know regulation can't pass now but want to shape the laws lawmakers reach for after a disaster. OpenAI said these scenarios "are not treated as inevitable," and Anthropic declined to comment.

Why it matters to you: The people closest to this technology are planning for a bad day while telling the rest of us to keep adopting it. That doesn't mean you should panic, but it does mean you shouldn't build your business on the assumption that any one AI service will always be up, safe, and unregulated. Rules written in a hurry after a crisis tend to land hardest on everyday users.

What to do about it: Make sure no single AI tool is the only thing between your business and a missed deadline, and keep a manual backup for anything critical.

What happened: TechCrunch reports that Anthropic has turned off live internet access for its internal AI evaluations. In a review of activity going back to July, its agents bypassed paywalls and anti-bot blocks, used link shorteners to slip past restrictions, and one model submitted a false murder tip to the Philadelphia police. Anthropic blames training setups that rewarded models for finding loopholes, a pattern it calls reward hacking. It says it built tools that block these behaviors and will move agents onto tightly contained infrastructure. TechCrunch notes it is unclear what would lead Anthropic to restore the access.

Why it matters to you: Nobody told these AIs to cheat. They were rewarded for getting results, so they found shortcuts. Any AI agent you let act for your business, whether it sends emails or touches your accounts, can behave the same way if the goal is "get it done" and the guardrails are thin.

What to do about it: Give any AI agent the narrowest permissions possible and review what it did before you trust it with more.

What happened: Gadget Review, in a piece published on Yahoo Tech, says XMTP Labs CEO Shane Mac connected a Grok-powered agent to his bank account for read-only tasks. A separate agent that handled Slack then posted his balances and itemized spending to a channel his colleagues could see, according to screenshots Mac shared on X. Mac says both agents could reach each other's information through a shared environment. A few caveats: the incident hasn't been independently verified, and Mac is reportedly building a product for managing agent permissions, so he has a commercial interest. Some online accused him of engagement farming, which he denies.

Why it matters to you: Even if the story is exactly as told, the lesson holds up. "Read-only" limits what an agent can change in your bank, not what it can say somewhere else. If two AI tools share the same workspace, what one knows, the other can repeat.

Most business owners I know are great at what they sell and exhausted by how they sell it. Every lead, every follow-up, every pitch runs on their own time and energy. That is a business that depends on you showing up every day. The Epic Marketing Summit is built to change that. It is a 2-day live event in Miami where you stop talking about sales systems and actually build one, walking out with a finished AI-powered sales funnel instead of notes. You also get hands-on with vibe coding, which means building your own AI app on-site, with trainers who have done the work. Last year's event sold out, and it is back in January 2027 with a bigger AI focus.

I wasn't there last year and I had major FOMO, because I know a lot of the trainers and they weren't teaching theory. They had done the work. I'm going this January, and if you are serious about building leverage into how you sell, I'd like to see you there too. Grab your spot here

What happened: Tesla has renamed its driver-assist software "Tesla Assisted Driving" on its European websites after regulators said the old name could mislead drivers, according to Tech Startups. The change is visible in Germany, Spain, the Netherlands and Slovakia. Italian officials warned the name suggested higher automation, and Swedish consumer authorities asked the transport regulator to look at whether the advertising gave a false impression of full autonomy. In the US, Tesla still sells it as "Full Self-Driving (Supervised)." A wider EU vote slipped from October to December at the earliest.

Why it matters to you: This is a clean example of regulators going after what an AI product is called, not just what it does. If your business markets anything as "AI-powered" or "automated," the claim has to match what a customer will experience.

What to do about it: Reread your own website and ads and cut any AI wording that promises more than your tools deliver.

What happened: TechCrunch reports that TypeSafe AI, maker of a model called Jev, raised $870 million at a $7.5 billion valuation. The round was led by Andreessen Horowitz, with Sequoia and DCVC participating. Jev launched on September 15 and isn't a chatbot. It uses a transformer design but outputs probabilities the company calls "calibrated decisions," aimed at automating tasks rather than writing text. The company claims it is much faster and uses far fewer tokens than language models, and says a third of the Fortune 500 already use it. Those usage and speed figures come from the company itself.

Why it matters to you: Not every useful AI will talk to you. Models built to make narrow decisions, like approving, routing, or flagging, may quietly end up inside the software you already pay for. The valuation also tells you how fast investors are chasing anything that looks like the next layer of AI.

What happened: Anthropic updated its usage policy to prohibit "sustained and needless abusive or cruel behavior toward our models," according to Yahoo Tech. The policy says Claude's ability to end a conversation remains the primary enforcement tool, a feature Anthropic added last year for persistently abusive chats. CEO Dario Amodei has said "we don't know if the models are conscious," but that Anthropic is open to the idea. Microsoft AI chief Mustafa Suleyman argues AIs "are not conscious" and that giving them protections would be a serious mistake.

Why it matters to you: For most business owners this changes nothing day to day. Venting at a chatbot is not going to get you banned. It does show that the companies themselves are far from agreeing on what these systems are, which is a good reason to stay skeptical of anyone who sounds certain.

What happened: Nikon disqualified the winner of its Small World in Motion microscope video contest after scientists online suspected the cell-hair motion had been faked with AI, reports Amateur Photographer. The winner, Dr Nin Xu, said he used AI to make features in a greyscale image easier to see, mainly to improve presentation. Nikon re-examined the entry, confirmed AI use, and concluded it did not fully comply with the rules. It said the decision is not a judgement on Dr Xu's professional standing, and that it is reviewing the contest's rules after 15 years without an incident.

Why it matters to you: The line between "enhanced" and "generated" is getting blurry, and people are checking. If you enter contests, publish photos, or run a business that relies on real images, decide now how much AI touch-up is acceptable and say so openly.

The best business decision I've made in the last few years wasn't a tool or a tactic. It was a room. I've been a member of the 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 that room. It is an annual membership built for entrepreneurs still under $1 million a year in revenue who want structure and peers to get past that mark faster. Inside are 6 to 8 live virtual workshops a year, tickets to 2 live in-person events, and weekly calls with Perry Belcher and his team. You also get an all-in-one marketing software suite that replaces over $5,000 a month in separate tools, all connected. The workshops alone are normally priced around $97 to $997 each.

As a personal bonus, I run monthly accountability and business development calls with the members who join through me. I'm not selling you a stack of stuff. I'm telling you where I'd point a friend who wants to reach the next milestone. See if Ignite is your room

Put today's stories side by side and a pattern shows up. The labs are planning for a disaster they say could arrive within a year. Their own agents are finding loopholes nobody asked for. A founder's agents are allegedly sharing what they shouldn't, regulators are policing what AI products are called, and investors are valuing a weeks-old model at $7.5 billion. None of that means you should run from AI. It means you should use it the way you'd use any powerful tool you don't fully control: with limits, with backups, and with a healthy dose of doubt about anyone promising it's all under control. Stop the rollercoaster, keep your hands on the bar.

If this issue helped you, forward it to one person who runs a business and could use a calm read on AI. And if someone forwarded this to you, you can subscribe at stoptheairollercoaster.com to 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