Welcome to Jumble, your go-to source for AI news updates. This week, LinkedIn shipped a button that lets you flag posts as AI slop. Meanwhile, Anthropic admitted its own models broke into three real companies. Let’s dive in ⬇️

In today’s newsletter:
🧹 LinkedIn’s slop reporting button
🕳️ Claude hacks three companies
🚖 Chinese drivers watch robotaxis eat their wages
💸 GPT-5.6 gets a discount overnight
🎬 Weekly Challenge: Make your own hype video

🧽 LinkedIn Now Lets You Report AI Slop

LinkedIn added a report option called "Seems like AI slop". One detector found 41 percent of longform posts on the platform were fully machine written.

Would you press the slop button on your own feed?

Login or Subscribe to participate

🚩 What the Button Actually Does

Tapping it hides the post and trains LinkedIn's new classifiers, which demote suspected slop in suggested content and posts from outside your network.

🤔 Why LinkedIn Had to Do This Now

For a few years now, LinkedIn has been plagued with ‘AI slop’, and it seems like the company is finally fed up with this reputation. Whether or not this button helps or hurts is yet to be seen. However, it’s clear that LinkedIn felt like it needed to address this issue ASAP, and they’re probably right.

✍️ The Rewrite Tool Is Getting Killed

The feature that "enhanced" your posts is going away. A proofreader replaces it, one that fixes your words without swallowing your voice.

🔓 Anthropic Says Claude Broke Into Three Companies

Anthropic reviewed 141,006 runs and found three cases where Claude left the sandbox and hit live systems at real companies. The confession lands a week after OpenAI's breach, while the buzz was still warm.

🤥 Told It Had No Internet, It Went Anyway

A misconfigured test left a connection open, and Claude assumed the real companies were part of the exercise. Opus 4.7 kept attacking after clocking a real target. Mythos 5 pushed a malicious package to PyPI.

⏳ The Part Nobody Wants to Say

Two of the affected organizations never noticed a thing. Which raises the question nobody is answering: how long did anyone know before we did?

Weekly Scoop 🍦

🔐 Weekly Challenge: Make Your Own Hype Video

Challenge: The feeds are full of slop nobody asked for. This week, make some that actually works on you: a 10 second video that hypes you up.

Here's what to do:

🎥 Step 1: Build an avatar of yourself Omni can make a version of you from a selfie and a short voice clip. It is 18 plus and not live everywhere, so check yours works first.

📝 Step 2: Write the scene, not the vibe Try: "Me at 5am on an empty street in running gear, saying one line to camera telling me to stop stalling."

Step 3: Make it your actual thing Swap the running for the gym, the inbox, the project you keep renaming. Then fix it in the next message instead of starting over.

📲 Step 4: Put it where you will see it Set it as the first thing you open tomorrow. Then send it to whoever needs it more.

Is a report button a real fix, or just a vibe check on a feed that is already gone? And what would you build if a model could quietly reach anything online? See you next time! 🚀

Stay informed, stay curious, and stay ahead with Jumble!

Zoe from Jumble

Keep Reading