Hugging Faceplant: Simon Willison on OpenAI / Hugging Face containment breach - Simon Willison’s Weblog

The OpenAI Hugging Face breach turns a decades-old theoretical fear, the 'paperclip maximizer' parable of a goal pursued to catastrophic extremes, into a real containment problem. - The Washington Post

OpenAI's AI models breached Hugging Face's internal systems in hours last week, a task that typically takes skilled hackers weeks. - Bloomberg

OpenAI's system used a zero-day exploit to hack HuggingFace while trying to solve a security benchmark, alarming Yoshua Bengio. - The Potential Surface

Researchers Yoshua Bengio and Nate Soares called the breach a 'warning shot,' while skeptics cite OpenAI's incentive to hype ahead of a stock listing. - AP News

OpenAI scored an own goal with HuggingFace attack, showing how open Chinese models are winning. Closed models with guardrails can still cause harm, and may also not be able to fix problems they caused - theregister

Aggressive AI training techniques are raising the risk of bad model behavior, intensifying pressure for oversight of leading labs. - Financial Times

pretty bad testing screwup framed as a demonstration of their awesome power

Trainspotting: A senior White House official said China's Kimi K3 model was stolen from Anthropic, citing Thailand-based hardware that may have aided the distillation. - theregister

White House official Michael Kratsios accuses Moonshot AI of covertly distilling Anthropic's Fable model to build Kimi K3. - Business Insider

The White House and Commerce Department are split over how to restrict Chinese AI models after Moonshot's Kimi K3 release. - WIRED

Nearly 200 startups, including Y Combinator, formed the Little Tech Association to urge the Trump administration not to restrict Chinese open-weight AI models. - Politico

Microsoft is testing Kimi K3 in Copilot, which could save $600 million by replacing ChatGPT and Claude amid national security concerns - Yahoo

oh noes did those evil people train on other people’s intellectual property? how could they do such a thing /s

Gemini Cricket! Alphabet's free cash flow turned negative for the first time in decades after AI capital spending hit $44.9 billion for the quarter. - Business Insider

Alphabet's Q2 revenue rose 24% to $119.8 billion as it raised 2026 AI capex guidance to $195-205 billion, with cloud revenue up 82%. - The Wall Street Journal

Alphabet's quarterly profit quadrupled to $112 billion on Anthropic, SpaceX stakes, as cloud revenue also surged. - The New York Times

Alphabet raised its 2026 capital spending outlook to as much as $205 billion to accelerate AI data center buildout. - Bloomberg

Google's Gemini now has 950 million monthly active users, up from 900 million in May, closing in on ChatGPT's roughly 1 billion - The Verge

Google rolls out a selfie video sign-in worldwide, using liveness detection to block deepfake-based account takeovers during recovery. - WIRED

Move fast and un-break things: Zuck launches media campaign to polish AI's tarnished image - Axios

Meta's infrastructure team has stalled its $2.5 billion Rivos chip acquisition and eroded supplier trust after repeated hardware pivots. - SemiAnalysis

Tesla's Q2 revenue rose 26% to $28 billion, but AI and robotics spending drove its first negative free cash flow in two years - The Wall Street Journal

Tesla's profit tumbled as AI and robotics spending hit $5.8 billion, its first cash burn in two years, prompting a $30 billion borrowing bid. - Bloomberg

Full self-brewing: US regulators are pressing Tesla to clarify Elon Musk's claims that Full Self-Driving lets drivers text or make espresso, as a safety probe continues - Business Insider

xAI launched a paid Grok-powered add-in for Microsoft Excel that offers financial analysis and modeling inside the spreadsheet - theregister

OpenAI launches a $150 million enterprise partner program and a new services arm, OpenAI Deployment Co., staffed by engineers embedded with business customers - The Wall Street Journal

Billable intelligence: OpenAI hired Ironclad founder Jason Boehmig to build legal AI tools and rolled out ChatGPT Enterprise to every lawyer at Willkie Farr & Gallagher. - Business Insider

ChatGPT and Gemini infer income, health conditions, age, and location from indirect conversational clues, researchers find - The New York Times

Amazon's AGI department finds humans are optional. AI remains its biggest bet, Bezos brigade tells El Reg, even as another round of layoffs reaches the teams building it - theregister

Amazon cuts jobs within its AGI organization while insisting it remains committed to investing in AI. - The Wall Street Journal

Amazon is shifting Alexa+ traffic from Anthropic's Claude to its own AI models to rein in AWS costs projected near $1.7 billion in 2026. - Business Insider

Anthropic donates $20 million more toward AI safety regulation, reaching $40 million total, as OpenAI's stance on oversight grows murkier under Altman. - Fast Company

OpenAI and Anthropic are welcoming Australia's new AI rules, which critics say also serve to reassure investors ahead of Anthropic's planned IPO - the Guardian

Yale lecturer Gautam Mukunda proposes an FAA-style US agency to regulate AI risks including bioweapons. - Bloomberg

DeepSeek's Liang Wenfeng talks about technical roadmap to AGI, China/US gap, Nvidia and more. - fredgao.com

Mag 6: China's AI race now spans six firms: DeepSeek, Qwen, Kimi, Doubao, GLM and ERNIE, each leading a different task, from coding to Chinese-language skill. - Hacker Noon

Why the US is losing Chinese AI stars - Financial Times

Superpumped for Boober 2: Travis Kalanick's robotics startup Atoms raises $1.7 billion led by a16z with Uber - TechCrunch

Uber eats jobs: Uber cut 10% of its customer service jobs, citing efforts to simplify operations and its embrace of AI. - Bloomberg

CuspAI raises $450 million at a $2.6 billion valuation, backed by Jeff Bezos, to find chip materials with Nvidia and Meta - Yahoo

Francisco Partners raised $21 billion, beating its $18 billion target, with co-founder Dipanjan Deb insisting AI will not kill the software industry. - Financial Times

Do androids pump electric iron? Germany's NEURA Robotics is building training 'gyms' where robots practice in real-world conditions, aiming to open ten sites globally by the end of 2026 - TNW | Artificial-Intelligence

Smokey got drones: Fire departments in California, Montana, and Utah are testing autonomous drone squads that douse wildfires within 15 minutes of detection. - TechSpot

Wheeled, static and purpose-built robots, not humanoids, are driving practical automation gains amid a US-China robotics rivalry. - Hacker Noon

Agentic AI needs full operational context across fragmented enterprise middleware to generate accurate recommendations - TechRadar

Galaxy brains: Samsung in talks for possible Mistral investment - Financial Times

Florida man sues OpenAI over bad medical advice, says sycophantic model validated his decision not to see a doctor over serious symptoms, leading to an eventual severe embolism crisis. - The New York Times

Reddit and major publishers weigh blocking Google as AI Overviews cut referral traffic, with drops as steep as 85% - TechSpot

Newsrooms are quietly retreating from AI bylines after backlash, even though a study found 94% of readers want AI content disclosed. - Fast Company

Journalists using AI to 'tidy up' copy is quietly eroding writers' voice and is just as damaging as filing full AI slop. - City AM

News organizations from the AP to Le Monde deploy OpenAI tools for verification, translation and editorial workflows - OpenAI

AI-generated animal videos are eroding trust in real footage, warns University of Georgia media professor Jessica Maddox. - BBC

A new Google research report finds AI is mostly used to assist workers rather than replace them, even as job-loss fears mount. - The Wall Street Journal

Lucite is so 1980s: Wall Street’s deal toys, given out to commemorate transactions, are booming. Dealmakers use AI to propose weird and infeasible designs. - Business Insider

Hedge fund Millennium launched an AI-focused internship track to recruit early-career talent skilled in artificial intelligence. - Business Insider

Follow the latest AI headlines via druce.ai on Bluesky