The people building frontier AI just asked the government to slow it down.
This week, 1,293 employees from OpenAI, Anthropic, Google DeepMind, and Meta urged Washington to pace AI research. The U.S. also banned Chinese humanoid robot imports, while Anthropic admitted it shredded millions of books to train Claude.
The race everyone calls unstoppable is the same one its own builders want to slow.
Where that tension lands, below.
Frontier Reads
Deep Current
Frontier Labs Ask for Brakes
1,293 employees from OpenAI, Anthropic, Google DeepMind, and Meta signed a statement urging the U.S. government to build safeguards that can deliberately slow automated AI research. The signers include Dario Amodei, Anthropic co-founders, and the chief scientists of OpenAI and Meta AI.
The concern is AI improving AI faster than oversight can keep up. No lab wants to slow down alone and lose ground, so they're asking governments to step in.
Within hours, OpenAI and Anthropic officially backed the statement.
No one wants to hit the brakes first. That's exactly why the letter exists.
Anthropic Shredded the Books
Court documents unsealed this week detailed "Project Panama," Anthropic's effort to buy, scan, and destroy millions of physical books to train Claude. Hydraulic cutters sliced off spines, scanners digitized the pages, and the paper was recycled.
A judge ruled that digitizing legally bought books counts as fair use, since it is a format change rather than a new copy. Anthropic separately settled the piracy portion, over seven million pirated titles, for $1.5 billion. The unsettling part is permanence: once a rare, out-of-print book is scanned and pulped, no copy survives. Read more→
U.S. Bans Chinese Robot Imports
The U.S. banned new imports of Chinese-made humanoid and quadruped robots, citing cybersecurity and supply chain risks. Since China produces about 85% of the world's humanoids, the move hits its biggest export market.
China called the decision "unilateral bullying" and signaled possible retaliation, including restrictions tied to rare earths and U.S. companies. Read more→
In Radar
Before we dive in..
I love when google AI takes weird questions seriously
China wins the cheap-model race in Asia. The U.S. is pushing a full-stack AI export program across Asia, but open-weight Chinese models running 60-90% cheaper are already winning cost-conscious buyers from Southeast Asia to American enterprises. Read more→
Brain waves for robot training. Startup Encord is testing headsets that read trainers' brain activity to tag moments of error and surprise, betting the real bottleneck in physical AI is scarce real-world data, not model design. Read more→
Grok ships a faster voice model. xAI's Grok Voice Think Fast 2.0 reasons while it speaks, hitting 0.70s to first audio and topping rival speech models on quality benchmarks, at $0.08 per minute. Read more→
Outside the feed
🌏 A creator shared an ad showing an AI agent collecting an overdue payment while the business owner sleeps, a simple reminder that AI is meant to automate repetitive work, not replace people.
🛩️ Susan Li (@SusanLiTV) posted worth-catching commentary in the AI-markets scroll this week.
One last thing: a viral AI clip of instruments playing themselves has been making the rounds. Obviously synthetic, still worth the ten seconds. Watch
