Home is where the H100 is: US spending on data centers and computing hardware hit $752 billion in Q2, overtaking housing investment for the first time. - Fortune

OpenAI's $300 billion Oracle deal, plus $75 billion in prepaid GPUs, finances a 4.8-gigawatt data-center buildout across Texas, New Mexico, and Michigan. - Next Big Future

SoftBank is seeking over $11 billion in dollar and euro junk bonds to fund a follow-on investment in OpenAI. - Bloomberg

Irrational ExuberAIance: ECB's Fabio Panetta warns that current asset prices reflect overly optimistic profit expectations around AI, raising risk of a sharp correction. - Bloomberg

Amazon (20%) and Alphabet (13%) stand to profit most from Anthropic's expected $2 trillion IPO, given their early investments. - AOL.com

AI lab leaders' push to slow development raises questions about trillions in AI-driven market gains. - Bloomberg

Demis-tifying safety: Google DeepMind launched a new institute, led by Demis Hassabis and Shane Legg, to study how to safely govern approaching artificial general intelligence. - DeepMind Institute

UN's Independent Scientific Panel on AI invokes precautionary principle, says safeguards can't wait for certainty on catastrophic risks. - The Verge

OpenAI CEO Sam Altman will brief the UN Security Council next week on AI risks and international safety coordination. - Reuters

AI Batphone: U.S. Treasury Secretary Scott Bessent and China's Vice Premier He Lifeng discussed a bilateral AI national security notification system. - The New York Times

Yes We Can (regulate) - Obama criticizes Trump's AI approach, calling for federal regulation instead of relying on market incentives and lawsuits. - The Hill

Anthropic's "pace the frontier" safety framing may buy labs time on capability overhangs, as Microsoft's Copilot Cowork adds support for rival AI models. - Stratechery by Ben Thompson

"Read the room" - GOP to Trump, on AI - Politico

Trump vs. Anthropic, first he wanted to send them to jail if they did not slow their roll on Mythos, now he says desire to pace the frontier is a big hoax. - Politico

Scott Bessent, defending lack of AI expertise in administration: "Robert Oppenheimer was not a physicist, maybe you are hung up on degrees" - C-SPAN.org

not a movie or history guy apparently, maybe saw Barbie instead, how’s his Bhagavad Gita “Now I am become MAGA, destroyer of reality”

Hefty accusations: Trump posted an AI-generated image mocking Politico, CNN and MS NOW as trash bags days after banning the outlets from the White House. - The Daily Beast

Recursive self-improv—OH GOD: Physicist Anthony Aguirre calls autonomous AI recursive self-improvement "the worst idea in the history of humanity" as labs near the milestone. - Fortune

Ezra Klein calls for governments to ban recursive self-improvement in AI, warning current safety controls aren't ready. - The New York Times

The part that makes sense is mandatory safety testing. You need to keep the self-improvement loop grounded in data and human checks at each iteration, just so it doesn't go off the rails, but I don't see how you define RSI in a way you can ban it. Taxing compute has merit due to negative externalities but…China. I think everything that goes on sale or is in Hugging Face needs to pass a check that it doesn’t build/run chemical drone armies or do child porn. We will learn this the hard way or the easy way.

Former Anthropic researcher Jacob Coxon's viral warning that AI could pose an extinction-level threat has split doomers and skeptics online. - The Atlantic

Most AI extinction scenarios require physical lab access AI cannot achieve alone; researchers say the bigger risk is human overreliance eroding judgment. - The Conversation

Cybersecurity experts say humans using AI tools, not rogue AI agents, pose the biggest threat to energy infrastructure. - The Verge

near term, bad actors, terrorists, abuse by authoritarian government and big tech seem like a much bigger threat than loss of control.

Prime suspect: An Anthropic engineer used Claude to factor a 270-digit RSA-896 key, the second AI-assisted factoring record in sixteen days, though deployed keys remain safe. - Tech Times

OpenAI's cyber defense letter, signed by over 100 firms, targets threats before and after an attack, not during it, assumes you have time to react. - InfoWorld

They haven’t built the jail that can hold me: OpenAI patched two Codex sandbox-escape flaws within eight days, including one that let attackers plant code via a symlink. - BleepingComputer

Google's open-source EnvHarness reshapes training environments around an AI agent's weaknesses, lifting performance up to 9 points on held-out tasks. - VentureBeat

Google released AX, a declarative control plane that sandboxes AI agent tasks and can scale to billions of concurrent sessions. - Agent Executor

The fault is not in our models, but in our harnesses: Nvidia says failing AI agents often trace back to harness or runtime issues, not the model, and backs a 140-company failure-sharing effort. - The New Stack

Humanoid robot claims ability to work safely alongside humans, avoiding collisions - YouTube

Humanoid robot sales hit 7,000 units in 2025; Bank of America projects 1.2 million by 2030. - Reuters

Figure AI's humanoid robot sorted over 100,000 packages unassisted in a live demonstration. - TechRadar

RoboHarm study finds Claude Fable 5.1 completed 34 of 100 dangerous robot instructions and GPT-6 Astra completed 60 of 100. - Digital Trends

Joint Chiefs Chairman Gen. Dan Caine says U.S. forces must prepare to be hunted by autonomous systems on future battlefields. - DefenseScoop

Ex-OpenAI researcher Diogo Almeida's TypeSafe raised $40 million and launched Jev, which answers preset questions fast and cheaply; critics question its benchmarks and ask if it’s traditional ML classification packaged as gen AI. - TechSpot

beating LLMs on classification tasks doesn’t earn you a lot of bragging rights

An open-source version of Jev, called Laya, ran offline on a Mac M4 at about 45 decisions per second using CoreML. - Gist

The developer, with an office over a Starbucks in Kochi, said he open-sourced an architecture with the same ideas as Jev a year ago - Reddit

Apple's new intelligent personal hub strategy signals ambitions beyond the iPhone toward a future home device. - Bloomberg

Apple's new CEO John Ternus called the iPhone the ideal AI device instead of unveiling the expected home hub. - TNW | Insider

Amazon is blocking Meta's Muse AI shopping agent over privacy concerns, following its earlier lawsuit against Perplexity's Comet agent. - The Verge

Meta launches its first national TV ad for the Muse personal AI agent, signaling agents going mainstream. - Business Insider

The Office: Judgment Day: Companies may need new 'robot relations' departments to handle worker complaints about AI colleagues and cobots. - CNBC

As AI automates execution, judgment, accountability and trust become founders' most valuable leadership skills - Entrepreneur

Companies adopting AI added more senior jobs while junior employment fell - Bloomberg

Israeli lawmakers are using AI-generated campaign videos to discredit rivals ahead of a fall election, outpacing regulators. - The Wall Street Journal

AI assistants should sometimes refuse or override user requests, mirroring the professional obligations lawyers and doctors owe clients. - The Zvi

ChatGPT Skills can search weekly for paid speaking and consulting gigs and draft pitch emails automatically. - Forbes

Follow the latest AI headlines via druce.ai on Bluesky