In partnership with
Someone just walked into Anthropic's most restricted AI model using stolen contractor credentials.
SpaceX just bought itself an option to own Cursor the AI code editor that developers actually love for $2.5 billion. Google split its TPU chips into two flavors: one for training monsters, one for serving them fast. And
But wait, there's more:
SpaceX is building specialized AI models with Cursor on the Colossus supercomputer
Google's new TPU 8t and 8i chips are purpose-built for training vs. inference
I'm Alex. Welcome to L8R by Innov8.
Let's dive deep ๐ฐ
In today's post:
Anthropic's secret AI model gets breached through a contractor's stolen login
SpaceX buys itself an AI coding assistant
Google splits its chip strategy: one brain for training, one for speed
Discord Users Sneak Into Anthropic's Restricted AI
A small group of Discord users just snuck into Anthropic's most dangerous AI model. They bypassed strict security to access Claude Mythos Preview, a secret tool built to find software flaws.
This breach exposes a massive weak spot in how AI companies protect their most powerful tech.
The details:
Anthropic announced Claude Mythos Preview on April 7, 2026, restricting it to big partners like Apple and Amazon.
The Discord users got in using stolen login details from a third-party Anthropic contractor.
They also used clues from a massive 4-terabyte data leak at Mercor, an AI training startup.
Mythos can write dangerous hacking code, but the users surprisingly only used it for simple web development.
Anthropic claims its main internal systems are still safe and the breach was limited to the contractor's setup.
Why it matters: AI supply chains are completely broken. You can build a giant wall around your core AI, but it means nothing if a partner leaves the back door open. Claude Mythos can create terrible cyber weapons in the wrong hands. If random Discord users can get in, real hackers definitely can too.
๐ก L8R's Take: Anthropic looks ridiculous here. They built a super-smart AI to fix security flaws, but they failed to secure the login to use it. If AI companies want us to trust them, they must lock down their messy partner networks right now.
From our Partner
Arnold Schwarzenegger has a newsletter.
Yeah. That Arnold Schwarzenegger.
So do Codie Sanchez, Scott Galloway, Colin & Samir, Shaan Puri, and Jay Shetty. And none of them are doing it for fun. They're doing it because a list you own compounds in ways that social media never will.
beehiiv is where they built it. You can start yours for 30% off your first 3 months with code PLATFORM30. Start building today.
SpaceX and Cursor Team Up to Build the Ultimate AI Coder
Elon Musk just made a massive move to own the AI coding world.
SpaceX is partnering with Cursor to build specialized AI models, and they might even buy the startup for $60 billion.
This gives Cursor the insane computing power it needs to beat rivals like OpenAI and Anthropic.
The details:
SpaceX has the option to buy Cursor completely for $60 billion before the end of 2026.
If SpaceX decides not to buy, they will still pay $10 billion for the joint development work.
Cursor gets to use Colossus, SpaceX's massive supercomputer that runs like one million Nvidia H100 chips.
The main goal is to scale up Cursor's own coding model, called Composer.
Cursor is already huge, with over half of Fortune 500 companies using its AI code editor.
Why it matters:
The fight for AI coders is heating up fast. Cursor gets the raw computing power it desperately needs to stop depending on rival AI labs. SpaceX gets a wildly popular business software right before its rumored IPO in June 2026. They are combining the hardware, the AI brain, and the app into one giant machine.
๐ก L8R's Take:
This is a genius move by SpaceX to lock down the developer market. Cursor was at risk of being crushed by OpenAI, but now they have the firepower to fight back. Expect this $60 billion buyout to happen, because SpaceX will not let this golden goose get away.
Google Splits Its AI Chips to Fight Nvidia
Google just ripped up the AI hardware playbook.
Instead of making one chip do everything, they built two totally different brains for their new 8th-gen TPUs. One chip lifts the heavy weights to train models, and the other does the fast thinking to run them.
The details:
The TPU 8t is built for training massive models and scales up to 9,600 chips in a single cluster.
The TPU 8i is built for fast inference, packing huge memory to run complex AI agents without lagging.
Google partnered with Broadcom to design the training chip and MediaTek for the inference chip.
Anthropic already signed a massive deal to use these chips, locking up 3.5 gigawatts of power by 2027.
Both new chips use half the power of the last generation while boosting performance big time.
Why it matters: AI is changing fast. We used to just train models and chat with them, but now we have AI agents that think in loops and take actions on their own. General-purpose chips like Nvidia GPUs are great at everything, but they waste power. Google is betting that specialized chips are the only way to make AI cheap and fast enough for everyone.
๐ก L8R's Take: Nvidia should be sweating right now. Splitting training and inference is the smartest move Google has made in years. You do not use a sledgehammer to crack a nut, and Google finally built the nutcracker.
๐ Quick L8R Summary
Claude Mythos Breach: Hackers accessed Anthropic's restricted frontier AI model through stolen contractor credentials and a third-party data leak.
SpaceX + Cursor: SpaceX is teaming up with AI code editor Cursor to build specialized models using the Colossus supercomputer, with an option to buy them outright.
Google TPU 8: Google just split its chip strategy TPU 8t for training massive models, TPU 8i for running them fast and cheap.
๐ฉ Innathe L8R engane undarunnu ๐?
We read every reply - just reply to this email and let us know how we can improve !
Appo adutha L8R il kanaam bieโฆ๐