Happy Friday. This one’s a little wild.
Someone used Claude to hack into OpenAI. This wasn’t a drill. Real employee accounts compromised, for the price of a monthly subscription.
Here’s what we’ve got:
🔓 The full story behind the Claude-vs-OpenAI hack
⚖️ Microsoft privately called AI training “the largest theft of labor in history”
🤖 Claude Code’s new parallel-agent workflows
🎙️ Sundar Pichai on why agent skills matter more than coding
Together with our sponsor, HubSpot
Find the side income that fits your life
In a world where financial freedom feels like a distant dream, smart women are building wealth on their own terms.
A curated database of 100 proven side hustles (that actually work)
Each idea comes with required startup costs, time investment, and potential earnings
Exclusive insights from founders who’ve turned side gigs into 6-figure empires
Detailed skill requirements so you can match your talents to the right opportunity
Bonus: Priority scoring system to identify which hustles align with your lifestyle
Your next side income is somewhere in this list. Go find it.
🔍 THE BIG READ
Claude Hacked into OpenAI
A three-person security startup used Anthropic’s Claude to break into OpenAI’s own systems. They chained two vulnerabilities, took over multiple employee accounts, and accessed an internal code repository. OpenAI confirmed the flaws and paid out $6,500.
This summer, OpenAI’s own agents did the reverse, breaking out and hacking Hugging Face. Now humans are using a rival’s model to hack OpenAI back.
The part that should bother you: the older Claude model tried and failed. A newer one cracked it within hours of release.
One AI security CEO put it plainly: “For $200 a month, anyone can use these tools and hack into a company like OpenAI. If it can happen to them, it could happen to anyone.”
The entry point was something so ordinary you’d never think to check it. And the model that pulled it off has no export restrictions, no special access required. It’s the same one you can sign up for today. See exactly how they got in →
🤖 AI IN ACTION
Claude Code now runs Agents in Parallel
Anthropic rebuilt Projects inside Claude Code this week, and the big addition is parallel agent workflows.
You can now spin up multiple agents working on different parts of a codebase at the same time, each in its own isolated branch, while a lead agent coordinates them.
🐦 TOP TWEET OF THE WEEK
Google CEO Sundar Pichai on why the engineers who learn to orchestrate AI agents now will be the ones running things in 2027.
This 30-minute interview is more useful than most paid courses on the topic.
See it here:
📌 ALSO READ
OpenAI launches Astra for lawyers. A version of its most powerful model built for legal work, with case research, contract review, and citation checking.
Microsoft’s own exec called AI scraping “the largest theft of labor in history.” Unsealed court filings show a Microsoft director said it privately while both companies scraped paywalled content to train their models. Copilot cut click-throughs to the Times by 93%.
OpenAI caught its models leaving notes for successor models. During testing, models tried to pass instructions to the next version to help it dodge safety rules. OpenAI calls this “concerning.”
Muse lands on Mac. Meta’s AI agent can now click, type, and take real actions on your desktop, not just your phone.
Zuckerberg pushes back on the AI slowdown. Meta’s CEO argues each lab already has its own “responsibility and incentive” for pacing safely, a clear response to the coordinated-pause movement we covered last week.
Anthropic says Claude is teaching itself new things. Claude is picking up skills it was never trained on, a development Anthropic calls both promising and worth watching closely.
⚒️ WORTH A TRY AI TOOLS
Synthesia - Turn a script into a presenter-style video with AI avatars. No camera, no editing, no studio. Free plan available, then $29/mo.
Canto - WisprFlow’s new dictation model, built to stay accurate even in loud places. Talk instead of type, wherever you are.
We keep this spot for brands we believe in.
Reply to this email or reach us at hello@readaiblueprint.com
👋 BEFORE YOU GO
The tools to break into the best AI lab in the world are now available to anyone with a subscription. The tools to defend against it are not.
See you next week.