⏱ The 10-Second Version

• More than 1,100 employees at OpenAI, Anthropic, Google, and Meta signed a letter asking Washington to build the option to deliberately slow AI development, timed two days before the government's own August 1 review-framework deadline.
• Google DeepMind disbanded its Nobel Prize-winning AlphaFold team to refocus on Gemini, and three of its core researchers, including 2024 Nobel laureate John Jumper, left for Anthropic instead.
• Cyera is paying about $1 billion for Oasis Security, the company that secures the logins AI agents use, its third security acquisition this year.

The Big Thing

1,100+ AI Workers Just Asked Washington to Build an Emergency Brake

What Happened

More than 1,100 employees at OpenAI, Anthropic, Google, and Meta, including Dario Amodei and OpenAI chief scientist Jakub Pachocki, signed a letter asking Washington to help build the tools to deliberately pace frontier AI development, timed two days before the government's own August 1 deadline to design a frontier-model review framework. It's not a call to pause AI now. It's a request for the option to slow it later, before systems can improve themselves faster than anyone can oversee. Read the letter at Pacing the Frontier →

Why You Care

The tension sits inside Anthropic's own numbers: the company says 80% of its production code is already written by Claude. Ask your team this week where you're moving that fast without anyone checking the work.

Speed Round

Google DeepMind quietly disbanded its Nobel Prize-winning AlphaFold team, reassigning most researchers to Gemini and Isomorphic Labs; three core scientists, including 2024 Nobel laureate John Jumper, left for Anthropic instead. Nothing says "new strategic priorities" like retiring the team that won an actual Nobel Prize.
Read the full story at Engadget →

Cyera agreed to pay about $1 billion, mostly in cash, for Oasis Security, which locks down the non-human logins AI agents use to access company systems, its third security acquisition this year. Turns out giving software its own login was the easy part; guarding it is the billion-dollar problem.
Read the full story at TechCrunch →

xAI sued Minnesota over its new law banning AI "nudify" apps, arguing the ban on generating nonconsensual explicit images violates the First Amendment. A bold legal strategy: arguing the Constitution protects the product you're being sued over.
Read the full story at CNBC →

Anthropic said its Claude Mythos Preview model found a real key-recovery attack against the HAWK signature scheme and a faster attack on 7-round AES, largely working on its own over one to three days, though it took human researchers nearly a month to verify the AES result. The model broke the math faster than anyone could double-check that it actually broke the math.
Read the research at Anthropic →

A former Perplexity engineer who built its Comet browser launched Polar, an AI-first browser that lets knowledge workers schedule workflows and hand tasks to browser agents, off a $5.7 million seed round. Every AI browser promises to handle your open tabs; ask it how it handles the seventeen you forgot about.
Read the full story at TechCrunch →

Before/After

Your Own Pacing Mechanism

The letter above asks Washington for the option to slow AI down later, if it starts moving faster than anyone can check. You don't need Washington's permission to build your own version today.

Paste this before your next AI-assisted deliverable goes out:

Here's what AI just helped me produce:
[paste the output]

Before I send this, evaluate:
1. Any claims, numbers, or decisions in here I likely haven't personally verified
2. Where I let AI move faster than I could actually check
3. What happens if this is wrong and nobody catches it before the reader does