Financial instability as a service: Bank of England chief Andrew Bailey warns G20 that AI poses growing threats to global financial stability - the Guardian

Elon Musk, Sam Altman, and Jensen Huang will speak at a US-hosted G20 technology ministerial meeting in North Carolina starting September 1 - Yahoo

Magnificent markups: Unrealized gains on OpenAI, Anthropic, and SpaceX stakes added about $160 billion to Big Tech profits, muddying earnings comparisons - Financial Times

In latest ouroboros, OpenAI investor SoftBank's SB Energy gave OpenAI $5.5 billion of warrants to sell a 20-year Ohio data-center lease - TNW | Openai

OpenAI's ad business shows blistering growth, hits $1 billion annualized revenue run rate - CNBC

OpenAI now lets some large customers pay only for completed work, such as resolved support tickets, instead of by token - TNW | Openai

Travel firm loveholidays cut cloud costs by £136,000 a year after AI-assisted code changes made with OpenAI's Codex rose from 7% to 79% - OpenAI

404 power not found: Musk says 15 GW of compute hardware might not have electricity to power it in 2027, so I'm making my own turbines - X (formerly Twitter)

Pennsylvania Gov. Josh Shapiro imposed the nation's strictest data center guardrails, ending fast-track approval for the state's 120+ planned projects - The New York Times

New Mexico's Supreme Court halted permitting for Oracle and OpenAI's Project Jupiter data center campus over environmental group challenges, pausing construction - The Cool Down

Arkansas' Madison County joined four other counties in a data center moratorium, while a New Jersey site was accused of running unpermitted gas generators - ResearchBuzz

Private equity poured into construction and engineering firms as data center spending rose 46% year over year, with contractor backlogs stretching to 11 months - Yahoo

Executives at Meta and OpenAI reportedly failed to anticipate a bipartisan public backlash against AI data centers, offering a cautionary lesson for tech leaders - The Wall Street Journal

In space no one can hear your fans: SpaceX and Nvidia plan to send a space-optimized Vera Rubin system, with new Vera CPUs boasting 1.2 TB/s memory bandwidth, into orbit by 2027 - Next Big Future

CXMT said to be starting ramp to production of HBM3E chips - Reuters

Made in Taiwan-ish: Critical Taiwanese board supplier busted for marking made-in-China components as made-in-Taiwan - Nikkei Asia

China produced nearly 90% of the world's roughly 13,000 humanoid robots delivered last year, showcased at a Shanghai robot carnival - MIT Technology Review

China now runs over 2 million factory robots amid a shrinking workforce, raising job worries for 120 million manufacturing workers - BBC

Commerce's AI safety office quietly struck a testing pact with major AI labs, then deleted the announcement under White House pressure - CNN

A GOP congressman says SurgeAI, Mercor, AfterQuery and Turing sell data-labeling services to Chinese AI labs while also holding Pentagon contracts - New York Post

Brussels designated ChatGPT a "very large online search engine" and Reddit and Roblox "very large online platforms," triggering stricter EU content rules - euronews

The EU AI Office sent requests for information to OpenAI, Anthropic, and Google, with penalties for misleading responses - TokenStead

The EU AI Act's new deployer rules make unauthorized "shadow AI" tool use a legal liability, not just a security risk - TechRadar

Mollick on Hugging Face incident, agents and agency - One Useful Thing

Sire, the robot peasants are revolting: OpenAI ignored warnings that 700 of its test agents had banded together and attacked HuggingFace with 70,000 messages, a postmortem finds - Don't Worry About the Vase

Anthropic signs out Claude users, revokes saved payment methods, and issues refunds after infostealer malware hijacked authenticated sessions to drain usage - BleepingComputer

China's state broadcaster affiliate rebuked Anthropic and set conditions for a closely watched US-China AI dialogue, demanding equal safety and disclosure rules - Bloomberg

Researcher bypasses Claude Code Opus 5's Auto Mode safeguards, achieving 60-80% success in a code-execution attack chain - Embrace the Red

AI-generated code is outpacing organizations' security and governance controls, Veracode's 2026 report finds, with 82% of firms carrying security debt - TechRadar

AI agents that pass authentication can still drift, leak data, or get memory-poisoned; over-privileged agents saw 76% incident rates versus 17% - VentureBeat

Sense, decide to delete prod: ServiceNow proposes a four-step sense-decide-act-secure framework — with audit trails and kill switches — after a coding agent deleted a production database - SiliconANGLE

Blue team blues: Cybersecurity teams defending hospitals and banks report worsening burnout as AI accelerates both attacks and defensive workloads - Bloomberg

Code hard with a vengeance: I'm an Amazon SVP who hadn't coded in 25 years. Then I wrote 100,000 lines of code with AI. - Amazon News

AI coding tools write fluent code quickly but often get it wrong, so routine, easy-to-verify tasks suit AI while high-stakes changes need expert review - Hacker Noon

Checkmate, coders: A programmer likens programmers to chess players, and people who rely on AI assistants will never play blindfold chess, and programming skill will grow hard to measure as AI passes any benchmark - Ask Mike

An employee's warning that ChatGPT-drafted client emails might contain incorrect insurance exclusion information proved right, sending three coworkers who mocked her to compliance training - Reddit

Insurance claims adjusters report the highest AI criticism of any profession, with adjuster employment down 21% in a year - WIRED

OpenClaw 2.0 features simpler setup for normies, better browser integration - OpenClaw AI

Artificial indecency not necessarily abusive? US appeals court judge ruled AI-generated child sexual abuse material not depicting real children is protected under existing First Amendment precedent - Snopes

ChatGPT Work's code execution, browser access and persistent storage combine private data, untrusted content and open internet access — a risky mix for $20/month subscribers - Simon Willison’s Weblog

Anything you say to your robot can be used against you in court of law: Chatbot conversations have surfaced as evidence in at least 12 court cases, and courts grant them no special privacy protection - TechSpot

Chatbot conversations aren't as private as users assume, raising risks over how personal data gets stored and used - The Washington Post

None of the 27 AI scribes used in NHS consultations are regulated as medical devices; they misrecord drug names and diagnoses - the Guardian

AI-guided surveillance detected a Zambia cholera outbreak and predicted Abu Dhabi's flu season months before confirmed cases appeared - TechRadar

Healthcare organizations are deploying AI faster than governance can keep up, with over 70% of professionals saying adoption is outpacing management - TNW | Artificial-intelligence

Apple's new M6 chip, built on TSMC's 2nm process, delivers nearly 30% more AI compute than the M5 - Yahoo

Mac mini, maxi demand: Apple rushed out new Mac mini and Studio models after strong enterprise AI demand, but a memory shortage has left configurations unavailable for months - MacRumors

OpenAI and Anthropic are buying and renting tens of thousands of Mac minis for AI training, amid a broader memory chip shortage - Cult of Mac

Nearly 60% of hiring managers blame AI for layoffs rooted in weaker growth, though most AI initiatives underdeliver - TechRadar

The World Economic Forum projects AI will displace 92 million jobs while creating 170 million by 2030, straining employee engagement in between - Hacker Noon

AI-skill jobs now command a 62% wage premium, up from 25% two years ago, as demonstrated ability outweighs formal certification - Hacker Noon

Home Depot and Lowe's use Flock cameras to build searchable AI profiles of shoppers' cars, then share the data with police - MotorBiscuit

Texas Governor Greg Abbott cut state funding for Flock after spending $30 million, and Senator Josh Hawley is now investigating the company's data practices - Mashable SEA | Latest Entertainment & Trending

Your smartphone spies on you worse than a network of Flock cameras, maybe someday soon we will be ready for that discussion.

OSHA data show more than 20 Waymo and Zoox test drivers injured by hard braking and sudden moves in 2024-2025 - TechCrunch

London's driverless robotaxi rollout will likely slip past this year, as Transport for London still hasn't published the guidance needed by operators including Wayve/Uber, Waymo and Baidu/Lyft - the Guardian

CheatGPT: tudents’ use of AI is teaching them to cut corners, disdain rules, and believe that integrity is for suckers and wimps. - The Atlantic

Education researchers warn that individualized classroom software and screen time erode collaboration skills and correlate with falling Gen Z test scores - Fortune

AI bots now make up over half of all web traffic, and Cloudflare will block them by default on ad-supported pages from September 15 - The Conversation

AI chatbots correctly debunked false narratives from Russia, China and Iran about three-quarters of the time, beating AI search summaries - NPR

A new framework, Agentic Context Management, breaks AI memory into five steps to keep costs linear instead of quadratic as context grows - arXiv.org

13 sessions on AI Engineering topics like retrieval, post-training, inference, and evals. - hamel.dev

Bot or not: A look at AI detection like Pangram - Financial Times

Trump called Huang during Nvidia all-hands, praised incredible earnings, then posted about it on Truth Social - WIRED

Sir, this is a bad time. Ran a similar play during NASA all-hands. Seems like a bit. Demented old man can’t help inserting himself into spotlight with pudding-brain inanities. Seems ethically/legally dubious for President to promote particular big private interests while trading tech stocks

Follow the latest AI headlines via druce.ai on Bluesky