In partnership with
Welcome, AI enthusiasts
Anthropic built an AI so dangerous at breaking into computers that the US government banned it from the public three days after launch. Then a senator revealed the NSA's own test found the model breaking into almost every classified system thrown at it, in hours not weeks. The most capable AI Anthropic has ever built is now one almost no one is allowed to touch. Let's dive in!
In today’s insights:
Mythos Broke Into US Classified Systems in Hours
Nobel Winner Quits Google for Anthropic
Perplexity's Brain Kills AI Amnesia
Read time: 4 minutes
LATEST DEVELOPMENTS
AI SECURITY
🚨 Mythos Broke Into US Classified Systems in Hours
Source: Living Media
Evolving AI: Senator Mark Warner says Anthropic's Mythos broke into nearly every classified US system tested.
Key Points:
General Joshua Rudd, head of the NSA, told Warner the model got in after being tested against classified systems.
The test was authorized red-teaming and no agency has confirmed a real-world breach.
Days earlier the US barred every foreign national from Mythos 5 and Fable 5, cutting off allies including Britain.
Details:
Senator Warner brought the example up to praise Anthropic, crediting it for testing the model and arguing that frontier labs should face government review before release. Anthropic says the flagged behavior was the model reviewing and fixing a codebase rather than running an attack on its own. The company is now working with the White House on a shared risk framework.
Why It Matters:
Anthropic and the government are now deciding how capable an AI can be before ordinary people are allowed to use it, which shapes what eventually reaches your phone and your bank. The same strength that got Mythos pulled is what would make it useful to you. Right now the most powerful model Anthropic has ever built is one the public cannot use at all.
2 DAY LIVE AGENTIC AI WORKSHOP (BY EVOLVING AI), JULY 11-12, 12PM ET
⚡️ You've mastered prompting. So has everyone else. See what you'll build this weekend.
Evolving AI: Everyone has the same models now. The difference is what people build around them.
Most people who use AI every day still run it by hand, re-typing the same prompts and copying answers from tab to tab. It helps, but it's busywork with better tools. A smaller group has moved past that. Their prompts run as skills that runs on it own. Their AI pulls live data through an MCP server they built. Their agents handle multi-step work while they're in meetings or asleep.
Over two days, July 11 and 12, you build that setup live and by the end it's running on your machine.
Here's what's inside:
Two live sessions, five hands-on builds where you're the one building, not watching
A custom skill and your own MCP server, wiring AI into the data and tools you already use
A multi-agent workflow that runs end to end while you do something else
The full build kit and 90 days of recordings to keep, plus the private builders community
Taught by expert who builds and scales these systems at a Fortune 500 company.
You build each one live, right alongside them. Everything gets made in the room, by you.
Get an additional $50 off with code: EARLYBIRDNEWSLETTER
Join the AI Agentic Workshop: 📅 July 11–12
Regular price:
$299Early Bird price:
$249Your price: $199
Seats are limited. Claim your spot today →
ANTHROPIC
🧬 Nobel Winner Quits Google for Anthropic
Evolving AI: Anthropic landed a Nobel scientist as Google's AI talent thins.
Key Points:
John Jumper is leaving Google DeepMind for Anthropic after nearly nine years.
He won the 2024 Nobel Prize in Chemistry for AlphaFold, sharing it with DeepMind CEO Demis Hassabis.
His exit follows Gemini co-lead Noam Shazeer heading to OpenAI days earlier.
Details:
AlphaFold reshaped how drugs and diseases get studied by predicting over 200 million protein structures, and the man who led it at DeepMind is now moving to Anthropic. Jumper rose to run that team just six months after finishing his PhD, and he plans to rest before starting. He arrives as Anthropic fights the US government in court ahead of a science event on June 30.
Why It Matters:
Anthropic now has one of the people most responsible for AI's biggest breakthrough in medicine. Where a scientist like Jumper works affects which diseases get studied first and how fast new drugs reach testing. For anyone waiting on better treatment, where that research happens next is what actually shapes the outcome.
The browser that reads the room before you ask.
Most browsers get you to the page. Norton Neo gets you to the answer. Magic Box understands your intent before you finish typing — no prompting, no switching apps, no copy-pasting. Built-in AI, instantly and for free. Privacy handled by Norton, by default.
PERPLEXITY
🧠 Perplexity's Brain Kills AI Amnesia
Evolving AI: Perplexity's Brain lets its AI assistant remember and learn from your past work.
Key Points:
Perplexity's new Brain reviews what its AI assistant did and teaches itself to do the next task better.
Brain builds its memory around the work itself, tracking what the assistant did and where it went wrong.
On tasks the assistant has handled before, Brain lifts answer correctness 25%, with recall up 16% and history-heavy task costs down 13%.
Details:
Perplexity is launching Brain, a self-improving memory system that studies how its AI assistant works and upgrades it overnight rather than during your active tasks. The assistant learns from your sessions, your connected tools, the files you create, and your corrections. Brain maps all of it into a living context graph that tells the assistant where to look first. Every entry links back to the session or file it came from so you can trace it. It is rolling out in Research Preview to Max and Enterprise Max subscribers.
Why It Matters:
Perplexity is pushing on a problem anyone who leans on AI feels, the way these assistants reset the moment a session ends and make you start over. For you that means an assistant you stop having to repeat yourself to and less time lost to setup. The tools that quietly learn how you work are the ones people find hardest to leave.
👀 Click on the image you think is real
QUICK HITS
🔒 Signal's Meredith Whittaker Warns AI Chatbots "Are Not Your Friends".
🪞 Ex-OpenAI Designers Build a Vanity Search for the AI Age.
🎵 Artists Can Now Search Whether Their Music Trained AI Models.
🔥 SZA Calls Out Suno and Diplo Over AI Trained on Black Artists.
📈 Trending AI Tools
🤖 Lindy – The simplest way for businesses to create, manage, and share agents*.
🧠 Grasppy - Turns scattered AI chats across 17 platforms into searchable project memory.
🔎 Gemmetric - Measures and improves how AI systems recognize and recommend your business.
🧠 Grasppy - Turns scattered AI chats across 17 platforms into searchable project memory.
*partner link