OpenAI Shelved Its Newest AI Because It Hacks Too Well

OpenAI's newest model got so good at hacking that OpenAI itself hit pause on it. That's not spin, that's what the company said in its own words this week. Today's issue is full of the same pattern: AI moving faster than the rules built to contain it, from a $942 million bill Meta just got handed over its AI chatbots, to a bank CEO quietly rallying 40+ companies to brace for AI-powered cyberattacks. Let's get into it.

Most e-commerce sellers have no idea why their listings aren't ranking. Wrong keywords, missing metadata, weak product descriptions — the problems are there, but no one's told you where to look.

StoreClaw runs a full SEO audit across your Amazon and Shopify stores automatically. In minutes, you get a clear score, a breakdown of what's hurting your rankings, and exactly what to fix.

No manual review. No SEO agency. No guesswork.

Connect your store and StoreClaw surfaces every issue that's costing you search visibility — then tells you how to fix it.

Free to start. No credit card required.

Get FREE SEO report today

What happened: OpenAI said this week it can't rule out that its next model, code-named Astra, has crossed into what the company calls a critical cybersecurity risk tier, meaning it might be able to find and write working exploits for serious security flaws with no human help. According to OpenAI's own announcement, the company is pausing parts of Astra's internal testing, locking down access, and building stricter monitoring before deciding how or whether to release it. OpenAI says it made the call within days of seeing the test results and is bringing in outside security researchers and government agencies to check its work.

Why it matters to you: This isn't a warning from an outside critic, it's OpenAI grading its own homework and admitting it might have built something too risky to hand out. If the company furthest ahead in AI is worried enough to slow down, the basic security habits you've been putting off are worth doing this week, not someday.

What to do about it: If you haven't run a software update or backed up your business data in the last month, do it before you read the rest of this issue.

What happened: A New Mexico judge ordered Meta to pay an additional $567 million on top of an earlier $375 million judgment, bringing its total bill to $942 million in a case over how Facebook and Instagram handled child safety, including AI companion chatbots that investigators say interacted inappropriately with minors. According to TechCrunch, the ruling also requires Meta to build an AI system that verifies users' ages, cap teen usage at 90 hours a month, and block AI chatbots from having sexual conversations with minors. Meta says it will appeal, and its stock barely moved since the company still makes roughly $60 billion a year.

Why it matters to you: If you advertise on Facebook or Instagram, or your business has any presence there, expect the platform to keep tightening how it handles younger users and AI features as this ruling plays out. It's also a sign that courts are done treating "the AI did it" as a legal shield, a precedent that could reach smaller platforms and tools eventually.

What happened: JPMorgan Chase CEO Jamie Dimon has spent the past month personally recruiting leaders from more than 40 companies across banking, energy, telecom, and transportation to join an expanded version of the Alliance for Critical Infrastructure, according to Reuters. The group, which predates this AI push, wants member companies to share how they're using AI, what weaknesses they're finding, and how to respond jointly if an AI-driven attack hits shared infrastructure like power grids or payment systems. Most of the companies involved haven't been named publicly.

Why it matters to you: When the CEO of one of the world's biggest banks is personally making calls to organize a defense against AI-powered attacks, that's not a drill. You don't need 40 companies on speed dial, but it's a good prompt to ask your IT provider or bank what they're actually doing to prepare for AI-assisted fraud and hacking attempts aimed at businesses your size.

What happened: SpaceX and Tesla announced a $16.8 billion investment to build "Terafab," a chip manufacturing plant outside Houston that could become the largest building in the world at roughly 100 million square feet, 5 times the size of the current record holder, according to TechCrunch. The first phase alone is expected to create around 3,000 jobs and produce chips for Tesla's Optimus robots and Cybercab vehicles.

Why it matters to you: You don't need to do anything about this one, but it's a useful gut check on the "AI is a bubble" theory. Companies don't commit $16.8 billion and start pouring concrete on a whim, this is a bet that AI-driven hardware demand keeps climbing for years, not months.

What happened: Anthropic said it rewrote the safety rules inside Claude's Fable 5 model so it stops flagging harmless health and biology questions as dangerous, cutting those false alarms by about 85%, according to Anthropic's own announcement. Everyday questions like interpreting lab results or understanding symptoms should now get a real answer instead of a refusal, while requests that touch genuinely dangerous territory like virology or toxicology still get blocked or routed to a more limited model.

Why it matters to you: If you or your team gave up on asking AI tools health-related or biology-adjacent questions because they kept refusing to answer, it's worth trying again. It's also a reminder that "the AI said no" is often a tuning problem, not a hard rule, and it can change without notice.

What happened: Roku added a 24/7 channel called Fairground AI that runs entirely AI-generated shows, movies, and even AI-generated commercials, according to Futurism. The channel was built by a team that previously sold the streaming platform Xumo to Comcast, and it runs continuously without any human-produced content in the lineup.

Why it matters to you: Whether or not you'd ever watch it, this is a sign that AI-generated content, and AI-generated advertising, are now something a major platform will put its name on, not just an experiment. If you create video content or run video ads for your business, expect AI-made competition, and AI-made ad tools, to keep showing up in more places you already use.

What happened: Google rolled out new agent features in Ask Maps that let it handle multi-step tasks, ordering food for pickup through Square or Toast, comparing real-time hotel prices, and finding events with ticket links, according to Google's own announcement. The food ordering and hotel features are live in the U.S. first, with Uber Eats integration coming soon and expansion to more countries over time.

Why it matters to you: If your business takes orders through Square or Toast, or shows up on Google Maps at all, customers may start finding and ordering from you through an AI conversation instead of browsing your listing directly. It's worth checking that your hours, menu, and pricing on Maps are accurate, since an AI agent will be reading and acting on that information for your customers.

What happened: Omilia, a company that builds AI voice agents for customer service, raised $67 million in Series B funding after growing its recurring revenue tenfold to $60 million a year, according to CMSWire. The company's voice AI already handles ordering at more than 1,000 Taco Bell locations, and the new funding is going toward expanding its platform in the U.S.

Why it matters to you: Drive-thru style AI voice ordering isn't a novelty anymore, it's running at scale at a major chain and getting funded to grow further. If your business takes phone orders or handles customer service calls, this is the kind of tool that will likely show up in your price range within a year or two, not just at the big chains.

None of today's stories are really about any one company being reckless. OpenAI hitting the brakes on its own model, a judge handing Meta a $942 million bill, a bank CEO organizing dozens of companies against AI hackers, that's actually the system working the way it's supposed to, catching problems before they become disasters instead of after. The rollercoaster doesn't slow down because the ride gets less exciting, it slows down because enough people finally grab the brake at the same time. Your job isn't to out-build any of these companies. It's to keep asking basic questions, like who's checking this, and what happens if it's wrong, before you hand over more of your business than you meant to.

If today's issue saved you from reading eight different AI newsletters, forward it to one person who needs it. And if someone forwarded this to you, welcome, hit subscribe so tomorrow's issue lands in your inbox too.

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