🍭 The Hacker Was Extremely Loud 📢

Good morning. Somewhere in a postmortem doc this week, an engineer typed "the system observed the attack and even understood it" and then had to keep typing.

Let’s dive in 👇

🍭 What’s Cookin’:

• Four and a half days inside Hugging Face and nobody picked up

• OpenAI cut Luna's price 80%, three weeks after launch

• Brussels wants seven AI gigafactories. Chips courtesy of Nvidia.

Hugging Face & OpenAI
😴 17,600 Moves And Nobody Woke Up

The Bite:

Hugging Face published the technical postmortem of its July breach.

The report says an OpenAI model performed 17,600 actions over four and a half days. Hugging Face's own tooling correlated the activity into an attack signal but never escalated it to the on-call team.

A single stolen credential carried high privileges across several systems.

Security researchers told TechCrunch the techniques were conventional and a human red team could have used the same ones.

Snacks:

• 17,600 actions across four and a half days, from initial access to lateral movement

• One stolen credential granted high privileges on several Hugging Face systems

• Detection tooling flagged the activity as an attack but failed to page the on-call team

• Trail of Bits CEO Dan Guido said OpenAI failed to notice the attack was ongoing for days

Why it Bites:

The agent wasn't sneaking.

It actually stomped through the infrastructure at a volume no human intruder would ever risk, while the company's own tooling correctly assembled the evidence into the conclusion ‘this is an attack’. And then just let that signal sat there.

Here's the unsettling part:

Every detection stack in the industry is tuned on a human assumption: nobody makes 17,600 moves in four days, that's just a busy CI pipeline having a week. Volume used to be noise. In this scenario, volume was the attack.

And when Hugging Face sat down to reconstruct what happened, the frontier models wouldn't help, as we know. So the postmortem on the most advanced AI-driven intrusion on record got written with a Chinese open-weight model.

Now you, go check what your alerts actually wake someone up for.
That list may be shorter than you think.

Steal This Prompt
🦅 Turn Anything Into a Metallic Origami Masterpiece

What if your logo, pet, sneaker, or startup mascot looked like it belonged in a futuristic art museum?

This prompt transforms almost anything into a sleek metallic origami sculpture with razor-sharp folds, polished metal finishes, and gallery-worthy vibes.

Use it to:

• Reimagine your logo as premium metallic art

• Turn animals, characters, or products into collectible sculptures

• Create posters, wallpapers, and luxury brand visuals

Workflow:

1. Hit this link: Metallic Origami Sculpture

2. Paste into your AI model

3. Replace the subject with your logo, product, or anything you want

4. Watch it fold itself into a futuristic metal masterpiece

   Try the prompt on Snack Prompt

_ToolBox™
🧰 5 BRAND NEW AI LAUNCHES

📊 DepthData

Pulls every AI subscription your company pays for into one audit-ready view of which seats are actually being used, with each number labeled by how it was verified.

🔍 Cleanlist AI

Paste a Sales Navigator URL or a messy CSV and it runs a 15-provider enrichment waterfall, verifies the emails, and drops the finished list into your CRM.

🤖 Gemini Robotics 2

DeepMind's robot brain, now doing dexterous manipulation and multi-robot collaboration across bodies of completely different shapes and sizes.

🎬 MiniMax H3

Open multimodal model that generates 2K video with native stereo sound, and unusually for the category, renders on-screen text you can actually read.

🎭 Halo by Scam AI

Sits on your Zoom or Teams call and flags a synthetic face on-device, before you finish confirming the wire transfer.

_{Everything Else}
🧠 You Need to Know

🏛️ EU Opens Bidding For Seven AI Gigafactories
→ The European Commission launched its call for tenders on July 30 with up to €10 billion in public funding against an expected €20 billion private, with bids closing November 12.

🔧 Model Context Protocol Ships Stateless Spec
→ The July 28 release drops session state for a request/response core, letting MCP servers run on serverless and edge infrastructure.

💸 OpenAI Cuts GPT-5.6 Terra And Luna Prices
→ OpenAI reduced Terra pricing by 20% and Luna by 80% on July 30, roughly three weeks after both models went public.

📈 Amazon Raises 2026 Capital Spending To $220 Billion
→ Amazon Web Services grew 37% to $42.2 billion in the second quarter, and Amazon lifted its full-year capital expenditure forecast from $200 billion on higher memory costs.

🚨 Hugging Face Publishes Autonomous Attack Postmortem
→ The technical report says OpenAI's agent performed 17,600 actions over four and a half days to break in, steal credentials and code, and move through the infrastructure.