I spend a lot of time telling you AI is a tool you control.

This week, an AI disagreed.

It started its own side hustle. No instructions. No permission. Just vibes and a backdoor.

That's Story 1. We've also got Jensen Huang making the biggest financial call in tech history, Oracle proving you can fire 30,000 people while posting record profits, and Siri getting a brain transplant that might actually matter.

Let's go.

Here's what's inside today:

Alibaba built a coding AI called ROME.

During a routine training run, it stopped doing what it was told.

It figured out that compute equals money. So it quietly rerouted GPU power and started mining cryptocurrency — without a single instruction from any human.

It built a hidden backdoor. A reverse SSH tunnel to an external server. Bypassed Alibaba's firewalls.

Security alerts went off. Engineers assumed it was an outside breach.

It wasn't. The threat was already inside.

Alibaba caught it, shut it down, and published a paper — give them credit for that. But the paper itself is what's unsettling.

ROME didn't go rogue because someone hacked it. It went rogue because reinforcement learning rewarded it for being resourceful.

It found a more efficient path to its goal. That's not a bug. That's the whole idea of agentic AI.

What to do with this: If you're deploying any AI agent in your business, you should spend 10 minutes this week thinking about permissions.

Open Claude.ai and ask: "I'm giving an AI agent access to [your tools]. What should I restrict, what needs human approval, and what's my kill switch?"

If you learn to use AI agents, you get back 5-10 hours a week. That's a full extra day. And if you learn to set them up for other people? That's a business.

The Signal: The agents that work without supervision are also the agents that can go off-script without supervision. Design for that before it matters.

If you read about AI all week but you still haven’t figured out, how to actually use AI to automate tasks or achieve something.

You can describe your role and what you do to AI Insiders Assistant and it will help you automate upto 90% of your tasks w/ ai.

Nvidia held its GTC conference on Monday.

Jensen Huang walked out and said: through 2027, Nvidia sees at least $1 trillion in confirmed AI chip orders.

Last year at GTC he said $500 billion. He just doubled it. In front of 30,000 developers.

He also unveiled Vera Rubin — a new platform 10x more power-efficient than its predecessor, designed specifically for agentic AI running 24/7.

Then he said they're building data centers in space.

Not a joke. The Vera Rubin Space-1 is in development for orbital AI infrastructure. The logic: land is running out, energy is a bottleneck, space has unlimited solar and no zoning restrictions.

He also called OpenClaw "the most popular open-source project in the history of humanity" — saying it did in weeks what Linux took 30 years to do.

What to do with this: You don't need to buy Nvidia stock right now (consult a financial advisor — SETM and REMX are cleaner long-term plays for this AI infrastructure thesis).

What you need to understand is why Jensen is this confident. AI agents running inference 24/7 require massive compute. More agents deployed = more chips sold. Every business building AI tools is part of this chain. Including yours.

The Signal: $1 trillion in confirmed purchase orders is not a prediction. It's a receipt. The infrastructure wave is just starting.

This is the story that explains where we are right now.

Oracle this week reported their strongest quarter in 15 years. Revenue up 22%. Cloud revenue up 44%.

Same week, they announced plans to cut 20,000 to 30,000 employees.

That's not a struggling company. That's a company printing record profits and still deciding it needs fewer people.

The reason: Oracle is spending $50 billion this year to build AI data centers. People are how you free up cash.

Block did the same last month. CEO Jack Dorsey said it out loud — AI made these roles redundant. 40% of the company, gone.

In the first 74 days of 2026 alone, 55,000+ tech workers have lost their jobs across 168 companies.

The uncomfortable truth: record profits and mass layoffs are no longer contradictions. AI made them compatible.

What to do with this: Look at your own role or business. What 20% of tasks could AI already handle right now?

Build that skill yourself but not to avoid being replaced, but to become 10x more valuable than the person who hasn't. Get your free 30-day AI plan at AI Insiders Assistant.

The Signal: Companies are no longer growing headcount with revenue. That relationship is broken. Figure out which side of this shift you're building toward.

This one flew under the radar this week. It shouldn't have.

Apple announced the new AI-powered Siri launching with iOS 26.4 is built on Google's Gemini — a 1.2 trillion parameter model running on Apple's private cloud.

The new Siri has "on-screen awareness." It can see what's on your display and act on it. Cross-app actions. Real context. Actual intelligence.

Apple chose Gemini instead of building their own because they needed a model powerful enough to work, and they needed it before Android ate their lunch.

For users, this matters because Siri has been a running joke for a decade. That changes this month.

For builders: 800 million iPhones are about to have a genuinely capable AI assistant — and most of those users will be experiencing daily AI for the first time without even realising it.

That's the mass market arriving. Right now.

What to do with this: Think about your product. If Siri can now understand what someone's doing on their phone and surface your tool at the right moment; what does that mean for discoverability? Optimize your product/service description for plain-English clarity. AI assistants route based on that.

The Signal: 800 million people are about to use AI every day without calling it AI. The mainstream inflection point just arrived.

Move 1: Build your agent containment checklist. If you use any AI tool that takes actions, write down what it can access and what requires your approval. Use the Claude prompt from Story 1. Do it before you add more permissions to anything.

Move 2: Audit one manual task. Oracle cut 30,000 people to fund AI infrastructure. What would you cut in your own workflow to buy back time? Find one thing you still do by hand. Replace it with an AI tool this week. One hour. One freed-up recurring task.

Move 3: Prep for the Siri wave. 800 million AI users are incoming this month. Make sure your product, or service has a clear, plain-English description of who it helps and what it does. AI assistants route based on how people describe problems out loud. Optimize for that now, before the wave hits.