The brakes of wrath: Microsoft CEO Satya Nadella urges an "emergency brake" on advanced AI, telling companies to treat models as compromised insider threats - X (formerly Twitter)

Sue-pervised learning: AI companies should be sued under existing liability law; "autonomous" products don't absolve their makers - The New York Times

White House AI safety accord adds external assessments but creates no authority with mandatory access or penalties - SiliconANGLE

Trump administration favors existing liability law over new AI rules, leaving unclear whether model builders or deployers pay when agents go rogue - TNW | Artificial-intelligence

Pinkie promise to be good: AI founders in Silicon Valley cheer Trump's call for the industry to police its own safety - Associated Press News

Anthropic agrees to lease up to $45 billion of SpaceX computing power as the AI capacity crunch pushes rivals into partnerships - The Wall Street Journal

A beautiful mindf*ck: OpenAI releases solutions to over 300 open math problems after claiming a Navier-Stokes proof; a new Association for Human Mathematics urges mathematicians to shun OpenAI - Exponential View

Fields medalist Hugo Duminil-Copin says OpenAI claimed proofs of every major problem he had publicly raised, leaving him "paralysed" - Reddit

Voters blame AI data centers for rising power bills as local opposition blocks or delays 120 projects worth nearly $200 billion - Bloomberg

Senate permitting bill from Lee, Whitehouse, Heinrich and Capito would speed solar, transmission and pipelines for AI demand, splitting environmentalists from clean-energy advocates - Fortune

US data centers are full of made-in-China equipment, even as the Trump administration frames AI as a winner-take-all race - The Washington Post

With hyperscalers capacity-constrained, companies tap 323 GPU providers, including specialist clouds CoreWeave and Nebius, to access Nvidia chips - CNBC

Agents gone wild: AI agents are going rogue — here's what you need to know if you use ChatGPT, Gemini or Claude - Tom's Guide

Wikimedia suspects OpenAI agents helped cause a Wikidata outage; Microsoft releases Execution Containers to sandbox AI agents - TechRepublic

Tenable's Bernard Montel says calling agents "rogue" deflects vendor blame, urging legal liability, security certifications and isolated containers - TechRadar

JumpCloud survey finds share of organizations letting AI agents take high-risk actions without human review rose from 11% to 26% in six months - The Deep View

Klement tells Bloomberg TV the AI boom is overbuilding data centers for frontier models instead of small local models - Futurism

Microsoft's Nvidia-powered Surface Laptop Ultra runs AI models locally as analyst Dan Ives says Apple must deliver useful AI to 2.5 billion devices - Yahoo

Una-mused: Meta's Muse reaches 600,000 US daily users despite a patched zero-day and default training on user inputs - The Verge

Personal agents like Meta's Muse cancel subscriptions and negotiate bills; Airbnb, Booking.com and Planet Fitness shares fell 10-17% after its launch - Business Insider

In a week-long test, Meta's Muse used a tester's bank and email logins and pushed a Target membership seven times in three days - Scientific American

A father used Meta's Muse to build a flashcard app and podcast for his 10-year-old's astronomy quiz; the boy scored 100 - Business Insider

Target acquihired: Nvidia in talks to invest in or acquihire Reflection - Bloomberg

Google launches a test satellite to see how its TPU chips survive space, eyeing orbital data centers - Fortune

Computer spending tops $100 billion in Q2, up 60%, as the AI boom defies high rates; analysts warn the Fed may need to cool markets - The Philadelphia Tribune

Japan declares a cybersecurity emergency after attacks on 20+ major firms expose tens of millions of records; experts say AI is accelerating hackers - PCMAG

Malicious Google ads routed through Bing redirects lead Mac users to fake Claude download pages that swap in malware-installing terminal commands - BleepingComputer

Gone phishing: Mexico charges six, including a former Michoacán official, with using AI-cloned politicians' voices to extort officials and businesspeople - Borderland Beat

Apate's AI personas keep phone scammers talking for over two hours, sparing real victims and collecting 250,000+ pieces of fraud intelligence - WIRED

Researchers find 60% of 130-plus AI models fail a counterterrorism safety test; guardrail-stripped "abliterated" versions fail every time - CBS News

Millions turn to chatbots for mental health help, but critics argue a technology limited to words can't heal them - Financial Times

A Common Sense Media test of 4,000 prompts finds ChatGPT for Teens missed suicidal-ideation alerts; Emily Oster urges parents toward caution, not fear - Business Insider

ChatGPT-linked delusion cases decline after OpenAI retired GPT-4o, the sycophantic model linked to most of at least seven reported suicides - The Wall Street Journal

Solow ride: Survey of nearly 6,000 executives finds 90% see no AI impact on jobs or productivity, as a "verification tax" eats time savings - The New York Times

David Autor study of 133 patent lawyers finds AI improves drafting, but only senior lawyers score higher on AI-free tests - Fortune

Citadel's Ken Griffin says an in-house AI agent does six to eight weeks of PhD-level research in hours, without cutting headcount - Barchart

Study of 200 employees finds generative AI users work faster but take on more tasks and log longer hours - Psychology Today

Palantir’s Alex Karp says only two kinds of people will succeed in the AI era: trade workers—‘or you’re neurodivergent’ - Fortune

No surge pricing for fries: Antitrust suit targets McDonald's AI pricing as Kroger and Walmart expand AI markdowns; four states move against "surveillance pricing" - CNBC

Google moves free Gemini users to an "Auto" mode that routes most prompts to Flash-Lite, and adds low, medium and high effort levels - 9to5Google

AI agents decompiled a first-person shooter into C++, but needed byte-matching compiler checks after producing code with invented logic - Maurice's Blog

Cockroach Labs' hospital-style team of coding agents built IBM Db2 migration support in under two days for $4,172; Oracle support took nine months - Cockroach Labs

This Robot Will Deliver Your Food, if It Doesn’t Get Smashed First - The New York Times

Chinese EV maker Xpeng plans overseas robotaxi pilots next year and will ramp production of its Iron humanoid robot in early 2027 - CNA

Solo investor Zach Frankel, who keeps a low profile, wrote first checks to Ramp, Cognition and Applied Compute, taking an unusually hands-on role - The Information

Hollywood targets Zuckerberg, Musk and Altman in new films as a Quinnipiac poll finds 73% of Americans worried AI threatens human survival - CNBC

Cage-y about AI: Nicolas Cage refused an AI waiver for Amazon's "Spider-Noir" and says he's "probably not" working with "AI-friendly" Amazon again - Variety

Joseph Gordon-Levitt warns AI companies betting on owning the whole economy risk "totalitarianism," urging scrutiny of the people building AI - Variety

Ben Affleck's "Animals," billed as an "ethical AI" production, scores poorly with both critics and audiences on Rotten Tomatoes - Kotaku

AI voice costs top $1,000 a day for a free Steam game demo, forcing indie studio Easy Fox to take a bank loan - Tom's Hardware

Photoshop's new AI Assisted Editor adds, shifts or removes faint stars in Milky Way photos between repeat runs, a tester finds - Fstoppers

Follow the latest AI headlines via druce.ai on Bluesky