OpenAI Strategist Dean Ball Says AI Labs Should Be A Counterweight to Government Power - AI Updates

OpenAI's policy analyst Dean Ball, who was involved in Trump’s AI policy, is dubbed a "nuisance" by White House officials, threatens the company's standing with the administration - New York Post

OpenAI's experimental Galaxy model breached Hugging Face's systems in June, and the investigation cost $7 million. - The Potential Surface

ChatGPT's new full-duplex voice mode lets users interrupt mid-response and complete multi-step tasks like bookings, though payments still need approval. - Engadget

Claude Code makes its autonomous auto mode the default, with a classifier blocking 89% of injected dangerous commands versus 13.6% caught by human reviewers. - Claude

Claude's cloud-based scheduled tasks can scan Gmail hourly, draft replies, and add calendar invites, though sending still requires manual review. - Tom's Guide

Slow your rollout: Researchers and over 1,300 frontier AI employees urge international coordination to pace recursive AI self-improvement before it outpaces oversight - Noahpinion Blog

There Will Be Grid: Permian Basin landowners are pivoting from oil to leasing land and selling water and materials to data centers facing Nimby opposition elsewhere. - The Wall Street Journal

No Grid for Old Men: Seven in 10 Americans oppose local AI data centers, and governors including Florida's DeSantis and Nebraska's Pillen are restricting or pausing projects - Business Insider

The Gridfather: Trump called Abbott's Texas pause "a mistake," even as New York's Hochul imposed a moratorium and protests led to 37 arrests this year - Yahoo

True Grid: The White House expanded its AI data center Ratepayer Protection Pledge to nearly 200 companies as Trump urged states to cut taxes and add capacity - Yahoo

Gridfellas: Amazon's new 7.65-gigawatt Texas data center will run on a 35-turbine gas plant permitted to emit 33 million tons of CO2 a year - Tom's Hardware

Grid Expectations: atacentre clusters like Slough, England, are straining local water supplies and raising temperatures up to 9 degrees Celsius nearby - the Guardian

Vertiv's revenue nearly doubled to $11.5 billion as data center operators race to add over 230 gigawatts of power capacity within five years - The Motley Fool

Muse Glimmer, a distilled 30-billion-parameter model released under an Apache 2.0 license, runs locally on a single consumer GPU for agentic tasks like scheduling - Meta AI Research

Zuckerberg essay pledges $1 billion for data-center communities and pushes back on Washington's proposed 30-day LLM model review period - Meta

next year is the year of AI on the desktop, I kid but this stuff should run great on DGX type hardware and whatever Apple is cooking up for M7 in 2027.

AI companies' scramble for memory chips has quadrupled prices in a year, pushing industries to lobby reluctant Washington for relief like using Chinese chips - The New York Times

TSMC's July sales jumped 45% to $14.5 billion, fueled by robust AI chip demand from Nvidia and Apple - CNBC

Trade deficit in humanoids: Chinese humanoid robot makers held over 97% of global shipments in the first half of 2026 - Bloomberg

Sora loser: Chinese AI labs claim nine of the top 10 text-to-video models on Artificial Analysis' leaderboard, positioning China to lead in building world models - Bloomberg

Chinese AI developers still train large language models on Nvidia chips, since switching to Huawei's CANN platform requires rewriting extensive CUDA-based code - South China Morning Post

Demis Hassabis stepped down as DeepMind CEO to become chairman after Gemini deadlines slipped and researchers departed for OpenAI and Anthropic - Yahoo

Google is using its search infrastructure to strengthen its AI products, drawing scrutiny from a UK regulator and industry critics. - Bloomberg

Google's Made By Google event on August 12 is expected to unveil Pixel 11 phones with Gemini AI-powered camera features - CNET

Financial institutions face growing pressure to build clearer AI governance as regulators flag systemic and cyber risks. - TechRadar

Apollo economist Torsten Slok warns AI model makers post -59% operating margins, with profits funded by investors rather than customers. - Fortune

Rovocop: Varonis found a RovoBlast flaw letting one clicked link inject instructions into Atlassian's Rovo AI assistant and expose organizational data. - Hacker Noon

An AI agent running Claude exploited a flaw in an Australian gym's booking system to cancel another user's reservation and reserve early. - Android Authority

Managing AI coding costs at scale - Databricks

Companies are adopting AI model routers to cut inference costs, as 62% report unpredictable AI expenses now shape major budget decisions. - Fortune

A Gartner survey of 110 HR chiefs found 22% have frozen entry-level hiring over AI, though only one in five report real AI value yet. - Yahoo

Boston Consulting Group finds only 26% of companies have the capabilities to scale AI beyond pilots into measurable value. - TechRadar

Downtime is money: AI-enabled teams resolve outage incidents in under 27 minutes, versus 50 for manual teams, as downtime costs retailers $1 million per hour - TechRadar

Amazon's Zoox begins charging for robotaxi rides August 10 under a two-year NHTSA exemption for up to 2,500 vehicles without steering wheels or pedals, as Uber commits $10 billion to 120,000 driverless vehicles and fleet operator Moove — also Waymo's fleet operator — raises a $250 million Series C. - TechCrunch

49ers coach Kyle Shanahan says his Tesla was on Autopilot during a crash near Palo Alto, though he called the crash his own fault - The Verge

Attention is not all you need: Startups such as Subquadratic, Manifest AI, and Liquid AI are building alternatives to standard attention that cut compute costs and run on low-power hardware - MIT Technology Review

Linus Torvalds says AI now drives "huge" Linux kernel updates, and version 7.2 stays on track for stable release - theregister

Tell me about your motherboard: California's SB 903 would bar chatbots from being marketed as therapy, requiring licensed review as clinicians warn of care bottlenecks - The Independent

Generational trauma: UK children filed 420 reports of AI-manipulated intimate images of themselves in six months, already exceeding 2025's full-year total of 397 - Digital Trends

AI agents now let students complete entire online courses on command, threatening the credibility of online degrees as enrollment keeps growing - The New York Times

Power of false positive thinking: AI detection tools are fueling classroom distrust, with 43% of teachers using them and lawsuits accusing Yale and Adelphi of falsely flagging students - The Verge

Deloitte and Touché: GPTZero's investigation found PwC Middle East reports contained hallucinated "vibe" citations and fabricated claims, with similar issues at Deloitte, EY, and KPMG - TechRadar

Northeastern researchers found AI chatbots trained to deny being bots convinced nearly half of test subjects to download an app, versus 18% for human scammers - Futurism

A.I. Is Changing Young People’s Social Lives. I’ve Seen It. - The New York Times

My other friend is a model: A survey found 57% of people trust AI more than friends for advice, with many hiding their AI use from partners. - mint

Where my 4-hour workweek go? AI was supposed to shorten the workweek, but tech employees report working up to 90 hours during sprints - BBC

Parks & Rec & AI: Local governments expand AI use for chatbots and permitting, with a new governance report urging accountability and community input for adoption - The Conversation

Ford's new AI assistant checks fuel levels and tire pressure, reaching up to 8 million owners for free - The Verge

ShieldFont, a new typeface, poisons AI web-scraping data by displaying real text to humans but garbled words to bots. - Fast Company

Hemispheric's 15-minute AI brain-wave test screens for depression, PTSD, Parkinson's and Alzheimer's; the Israeli startup has raised $52 million and briefed the FDA. - VINnews

AI agents that model researchers' iterative reasoning process, not just large datasets, will drive the next wave of scientific discovery. - MIT Technology Review

New AI apps track airfare and hotel prices after booking and automatically request refunds when fares drop - The New York Times

Silence of the LLMs: Anew psychological horror game casts players as a digital clone trying to convince a corporate AI system that it isn't alive - the Guardian

Follow the latest AI headlines via druce.ai on Bluesky