In partnership with

Hello again, human brain, and welcome back to your daily munch of AI news.

Here’s what’s on the menu today:

• Your AI security is a joke🤦‍♂️

  Hackers broke through Claude's safety system.

• Meta stole 81.7 TB of books🦹‍♂️

  Tsk tsk tsk, Zuckerberg…

  New here? Subscribe! 😎

Peep today's Spot the AI at the bottom. 👇

MAIN COURSE

AI safety = Swiss cheese🤦‍♂️

On Thursday, Anthropic shared the results of its latest jailbreaking challenge. Hackers broke through all 8 levels of Claude's new safety system, despite claims it was “robust” against jailbreaks.

What happened?

Anthropic shared the results of its latest jailbreaking challenge. 4 hackers beat all of Claude’s security levels in 6 days. One discovered a universal jailbreak, which acts as a master key to bypass all safety limits.

Give me the deets:

• 13,960 participants

• 800,000+ chat attempts

• 10,000+ testing hours

• $55,000 in total prizes paid

So... How can I hack AI?

The hackers bypassed safety guardrails by:

• Ciphers to hide bad content

• Role-play scenarios

• Swapping dangerous words for safe ones

• Prompt injection attacks

• and more.

What's Anthropic saying?

Safety classifiers help, but aren't enough. AI models need multiple layers of protection. It’s time to go back to the drawing board to improve defenses before releasing more capable models.

Ready to level up your work with AI?

HubSpot’s free guide to using ChatGPT at work is your new cheat code to go from working hard to hardly working

HubSpot’s guide will teach you:

• How to prompt like a pro

• How to integrate AI in your personal workflow

• Over 100+ useful prompt ideas

All in order to help you unleash the power of AI for a more efficient, impactful professional life.

Get the free guide and level up your AI game today!

SIDE SALAD

Zuck’s theft exposed🦹‍♂️

Two weeks ago, court records revealed Meta employees downloaded massive amounts of pirated books for AI training. This occurred despite numerous internal warnings about copyright violations.

What happened?

Meta staff downloaded 81.7 TB of books from pirate sites like Z-Library and LibGen to train LLaMA.

Who knew?

Who didn’t? The staff said, “Using pirated material should be beyond our ethical threshold.” Zuck's response? “We need to move this stuff forward… find a way to unblock all this.”

Is this common?

OpenAI, Nvidia also sued for using pirated books and videos in AI training. DeepSeek now accused of stealing from ChatGPT.

What's the fallout?

There’s a class action lawsuit ongoing. With Meta's deep pockets, the legal battle could drag on for years. The court will decide if Meta directly infringed copyrights, but expect a long appeals process.

YOUR DAILY MUNCH

Cool Tool 🛠️

Is your life insurance keeping up with your life?

• Getting married or having kids changes financial needs.

• Make sure your life insurance policy reflects this.

• Comparing plans often helps ensure you’re appropriately covered

View Money’s Best Life Insurance list to find coverage for as low as $1/day.

Think Piece 🧠

All of America’s data in 1 AI? Larry Ellison advocates for all US data, including DNA, to be in a single Oracle database to streamline healthcare, food production, and more.

Startup News 💰

Elon Musk demoed Grok 3. It’s the latest AI system from xAI, trained on Tesla's “Colossus” supercomputer. Musk claims it’s the “smartest AI on earth.”

Research 👨‍🔬 

The “stochastic parrot” — AI memorizes but fails to genuinely understand concepts. It restates physics principles but fails to interpret them when presented in unfamiliar formats.

InfiniteHiP — repurposes diffusion transformer attention layers to create sharp, interpretable saliency maps that precisely locate textual concepts in images.

MEMES FOR DESSERT

SPOT THE AI

3 of these are real horses. 1 is fake. 🐴

Which one is AI-generated? 👇

Ideas? Comments? Complaints?

Respond to this email or hit me up on 𝕏.

Until next time 🤖😋🧠