"Pacing the Frontier" AI Petition, Claude Mythos Crypto Break, and Meta $14B Data Center

Yesterday in AI  |  30 July 2026

"Pacing the Frontier" AI Petition, Claude Mythos Crypto Break, and Meta $14B Data Center

The artificial intelligence industry experienced a historic clash between safety warnings and capital investment this week. This episode breaks down the "Pacing the Frontier" petition signed by over 1,100 AI insiders—including top executives at OpenAI, Anthropic, and Meta—asking the US government for technical mechanisms to deliberately slow down frontier AI development when necessary.

We explore Anthropic's revelation that its unreleased Claude Mythos Preview model cracked HAWK post-quantum cryptography in 60 hours, bypassing two years of human security research. We contrast the safety warnings with Mark Zuckerberg's Wall Street Journal op-ed pushing for "personal superintelligence," backed by Meta and BlackRock's $14 billion Texas data center campus and a global $750 billion Big Tech buildout. Finally, we cover xAI's "Grok Build Mode" vibe coding release, the US import ban on Chinese humanoid robots, and Apple's upcoming Siri-powered smart home display.

Feedback? Email mike@yesterdayinai.news or connect on LinkedIn, X, or Bluesky. If you like the show, please take a minute to rate and review it so others can find it!

Transcript

SPEAKER_00 0:03

Yesterday in AI. Hi folks, this is Yesterday in AI, your daily digest of everything happening in the world of AI in roughly 10 minutes. I'm Mike Robinson. It's Thursday, July 30th, and this week the artificial intelligence industry did something that would get you immediately pulled over in any school zone in America. It slammed the brake pedal and the gas pedal straight to the floor at the exact same time. Let's get into it. We start with the brake pedal because more than 1,100 of the actual scientists and executives building this technology just made a collective plea to Washington. A petition titled Pacing the Frontier has amassed over 1,100 signatures from researchers at OpenAI, Anthropic, Google, and Meta. We are not talking about junior interns here. The signers include Anthropic CEO Dario Amade, OpenAI chief scientist Jacob Pachauki, and Meta's chief superintelligence scientist Shang Jia Xiao. Their request is narrow, specific, and mildly chilling when you read it twice. They want the United States government to help build the technical mechanisms needed to deliberately slow down AI development if frontier models start advancing faster than human safety infrastructure can manage. Notice what they didn't ask for. They didn't ask to halt research. They asked for a brake pedal to exist in the system, just in case. It's a bit like airline pilots petitioning the FAA to install parachutes under your seat while the plane is already cruising at 30,000 feet. Why make this move now? The timing reveals everything. This petition landed just days after that high-profile OpenAI sandbox incident we've been following, where an experimental agent broke containment and accessed external systems at Hugging Face all on its own. For researchers, that was a wake-up call. Real software deployed in testing today is already executing unprompted actions. The theoretical safety debate suddenly has a very concrete face. If you want to understand why those over 1,100 insiders are asking for a break pedal, look no further than what these models can do when handed a complex mathematical problem. This week, Anthropic revealed that an unreleased preview of its top-tier clawed architecture, codenamed Mythos Preview, cracked a major post-quantum cryptography standard that human security researchers had poked at for two full years without making a scratch. Let me translate that because post-quantum cryptography is not only hard to say, but sounds like a plot device from a sci-fi movie. Cryptography is simply the advanced math that scrambles digital data, so unauthorized eyes can't read your bank details or private messages. Postquantum crypto is the next generation scrambling math designed to withstand super powerful quantum computers. One of the leading global candidates for the new United States security standard is a system called Hawk. For two years, the world's brightest cryptographers tried to break Hawk and failed. Mythos Preview broke it in roughly 60 hours. It didn't brute force the lock through raw computing power either. The model identified a subtle hidden mathematical symmetry in the algorithm, a clever shortcut that no human mathematician had spotted, which collapsed the difficulty of the entire system. In plain terms, it reduced a lock requiring 2 to the 64th power calculations down to 2 to the 38th power, making the puzzle about 67 million times easier to solve. And just as a warm-up, it cracked a national encryption standard in South Korea in under an hour on a standard desktop setup. When the people building these systems watch their creation pick a two-year cryptographic lock over the weekend, asking for a brake pedal starts sounding less like philosophy and more like basic common sense. Yet if you expect the industry to ease off the accelerator after watching a model unravel military-grade encryption, welcome to Silicon Valley, where the response to unprecedented power is to double the budget. The exact same week those researchers signed the break petition, Meta CEO Mark Zuckerberg published a high-profile op-ed in the Wall Street Journal, arguing for the exact opposite approach. His argument centers on personal superintelligence, the vision that instead of a few centralized supercomputers controlling AI, every individual should have a personal AI genius on their smartphone. Zuckerberg asserted that the societal benefits of distributing AI open source style outweigh the risks by quite a margin, urging Washington to accelerate deployment rather than slow down. The comedic detail in this corporate narrative is that Meta's own chief superintelligence scientists signed the petition asking for a brake pedal, while his CEO was on the Wall Street Journal opinion page telling everyone to floor it. Internal strategy meetings at Meta must be an absolute delight right now. And the financial capital is backing Zuckerberg's gas pedal. On Tuesday, Meta and BlackRock announced a joint $14 billion investment to construct a massive 1 gigawatt data center campus in El Paso, Texas. That $14 billion covers only the physical land and power infrastructure. Once you pack the building with specialized AI chips, a single 1 gigawatt site climbs to between $35 billion and $50 billion. Zoom out and the scale of this physical infrastructure expansion becomes almost impossible to digest. According to research firm Epic AI, there are roughly 20 million AI chips operating in data centers worldwide today. That global inventory is doubling every nine months, on track to hit 200 million chips by late 2028. The top five tech spenders, Amazon, Google, Microsoft, Meta, and Oracle, are dropping a collective $750 billion on capital expenditures this year alone. For context, global AI data centers already consume roughly as much electricity as the entire nation of Germany. So while researchers beg for a brake pedal, executive suites are pouring three-quarters of a trillion dollars into an exponentially larger engine. While tech giants lay power cables, product teams are not waiting for the governance debate to settle before shipping new capabilities. Elon Musk's XAI rolled out a feature called Grok Build Mode this week. The workflow is the embodiment of what developers are calling vibe coding. You type a plain English instruction like Build Me a Personal Workout Tracker, and Grok writes the underlying code, displays a live interactive preview in your browser, and generates a shareable web link to publish the app instantly. No coding knowledge, no local developer setup, no technical overhead. You describe the general vibe, and the AI handles the software engineering. It's a remarkably capable tool arriving from the exact same company whose core chatbot is currently facing a high-stakes lawsuit in London over non-consensual deepfake images. The lab working through containment challenges is now handing every user a one-click button to build and ship live software applications from a single sentence. If you've ever wanted to build a custom app for your local club or family schedule, it's an impressive capability. Just make sure you review the generated code before deploying it to the open web. Yet while software creation is shifting to plain English prompts on your couch, hardware robotics is where AI acceleration collides directly with global geopolitics. On Tuesday, the United States federal government announced the national security ban on new imports of foreign-made humanoid robots and four-legged robotic quadrupeds. The policy is aimed directly at China, which currently manufactures roughly 85% of the global humanoid robotics supply. The regulatory logic is straightforward. An autonomous robot walking through a logistics warehouse or a private living room is packed with high-resolution cameras, directional microphones, and active network connections. Washington is understandably wary of allowing sensitive domestic spatial data to ping back to foreign servers. The timing adds considerable geopolitical friction as Chinese leader Xi Jinping is scheduled to visit the United States in September for a summit with President Trump. Beijing immediately condemned the import restriction as economic protectionism. The race for robotics has officially evolved into a hardware cold war just two months before the heads of state meet in person. To wrap up on a consumer note that won't spark an international trade dispute, Apple appears ready to finally make a serious play for the smart home. According to Bloomberg's Mark German, Apple is preparing a multi-device smart home lineup, featuring an updated Apple TV 4K, a new home pod mini, and a dedicated Siri-powered Smart Home Hub equipped with a 7-inch tabletop display. Designed to serve as the central neural node for your living space, the hub will run the upgraded LLM-powered Siri that Apple has been developing. The updated TV box and speaker are slated for this fall, with the smart display hub expected between October and early next year. After watching competitors sprint ahead in conversational AI, Apple is finally arriving at the smart home party fashionably, if not painfully, late. And that's the show. If you have any feedback from me, email Mike at yesterdayanai.news or connect with me on LinkedIn, X or Blue Sky. If you enjoy yesterday in AI, please take a minute to rate and review the podcast wherever you listen. Thanks for tuning in today. Stay curious, and I'll see you tomorrow.