CrowdStrike reports an 89% surge in machine-assisted cyberattacks as patch windows shrink to 48 hours. - theregister

Welcome to the Breach Club: Hugging Face CEO Clem Delangue called for mandatory disclosure laws covering AI agent cyberattacks after the OpenAI breach. - Business Insider

The Genie is out of the Bottle. - Schneier on Security

Zvi Mowshowitz argues cyber incidents expose failed alignment training and monitoring, warns similarly capable open models are coming within a year. - The Potential Surface

A threat actor switched to DeepSeek for autonomous hacking after Claude and OpenAI blocked the attempt, but DeepSeek's attacks failed due to missing technical prerequisites. - Tech Times

AI-assisted hackers could alter DNA evidence files without traces - The Wall Street Journal

Alibaba released Qwen3.8-Max, a 2.4 trillion-parameter model it says matches Anthropic's Fable 5 on several benchmarks. - Bloomberg

US startups like Arcee AI race to build open-weight AI models rivaling Chinese systems such as DeepSeek and Qwen despite thin venture funding - The Wall Street Journal

Spy hard: Anthropic told Congress that Alibaba ran the largest known distillation attack on Claude, using millions of prompts from fraudulent accounts to extract AI capabilities. - Forbes

OSTP director Michael Kratsios's distillation claims against Moonshot lack evidence, and experts say autonomous AI hacking may be the bigger threat. - Fast Company

A Reuters-Jamestown review found Chinese military-linked institutions distilling US models, including GPT-3.5 and Claude, for defense applications. - SiliconANGLE

Silicon Valley is split: Nvidia, Microsoft, and Google back open access to Chinese models while Anthropic's Dario Amodei urges restrictions. - Rest of World

DeepSeek's new V4-Flash model runs tests for $0.03 on average, far cheaper than rivals Kimi K3 and GPT-5.6 Sol. - Reuters

The Siri Re-awakens: Apple's Siri AI launches this fall in iOS 27, searching personal data and the web to challenge ChatGPT with emails, texts and images. - Bloomberg

CapExodus: Warren Buffett likens AI capex to the 19th-century railroad boom as hyperscalers eye up to $4.5 trillion in spending by 2030 - TechRadar

Economist Howard Rubin says companies are burning through AI token spending without established accounting practices to measure returns - The New York Times

A new industry of AI cost-cutting coaches, measurers, and builders emerges as one startup CEO calls enterprise AI spending "widely irrational and inefficient" - Business Insider

Kid CUDA: A startup used AI coding agents to recreate CUDA-like chip software in 10 hours, fueling debate over Nvidia's software moat. - Business Insider

The MI355X chip beats Nvidia's B300 on performance per dollar running Kimi K3, since B300 costs about 2.4x more despite leading on raw throughput. - Wafer

Majestic Labs unveils a server swapping Nvidia GPUs and pricey HBM (high-bandwidth memory) for Arm/RISC-V chips and up to 128TB of cheaper LPDDR6 memory. - TechRadar

Acrab says its Agent Box can approach Nvidia DGX Spark performance for one-fifth the cost and half the power - TechRadar

Nvidia DGX Spark could run quantized DeepSeek V4 Flash. - MicroCenter

Watermark this: California's AI Transparency Act took effect August 2 with fines starting at $5,000 a day, and Midjourney has yet to ship a required watermark. - Tech Times

New EU AI Act rules require chatbots and AI-generated media to disclose they aren't human, with fines up to 7% of turnover. - TechRadar

The proposed CHATBOT Act would mandate parental monitoring accounts for teens' chatbot conversations, raising privacy and age-verification concerns - Electronic Frontier Foundation

Bipartisan majorities of Americans back a federal AI oversight agency, mandatory testing, and deepfake labeling - Miami Herald

As generic AI copilots become commoditized, industry-specific vertical AI is emerging as the real differentiator, with Oracle citing 15-20% productivity gains in European workshops - TechRadar

AI tools now generate infrastructure-as-code faster than operations teams can govern it, creating configuration drift and rising cloud costs - TechRadar

Bad bot: Unitree unveils its AS2-W robot wolf, a four-wheeled platform reportedly using Nvidia AI hardware and already drawing interest from China's military - TechRadar

James Cameron and Arnold and Black Mirror warned us about this stuff

Foundation Future Industries tests its Phantom humanoid robot for industrial and potential military uses amid accountability concerns - The Independent

An unsecured Chinese police dashboard tracked nearly 12,000 foreigners, including journalists, via surveillance cameras and facial recognition, researcher Marc Hofer found. - The New York Times

Nanit's AI-powered baby camera logs infants' sleep and movements for 1 million users, topping $100 million in revenue as it fights a surveillance-tech image. - The New York Times

DuckDuckGo sells $35 camera-free sunglasses to protest privacy-invasive AI glasses - CNET

Gov. Newsom's office says it worked with YouTube to remove over 700 accounts spreading AI-generated lies about businesses fleeing California. - Semafor

AI-generated images and video are flooding social platforms, eroding people's ability to tell what's real. - The Wall Street Journal

Fraunhofer's RealorRender reports 85% to 91% accuracy detecting deepfakes - New Atlas

Lib Genesis: Booksellers in Australia and other book suppliers report abnormally large, price-insensitive orders for rare titles, suspecting AI firms are stockpiling training data outside copyright limits. - the Guardian

Computer science enrollment fell over 8% this spring as colleges add AI-literacy requirements for all majors - Yahoo

A PwC survey of 1,000+ US finance executives finds 86% say AI training beats an MBA for new hires - Bloomberg

Retraining for a new AI-era career won’t get you a job, employers demand experience in a catch-22 - The Economist

AI-assisted 'vibe coding' lets non-programmers build real tools, shifting expertise up the value chain instead of replacing developers - New Atlas

AI is changing how people get hired more than whether they keep a job, weakening resume signals for entry-level workers, experts say - Yahoo

UNAM voids 3,000 online admissions tests after a cheating scandal exposes AI-monitoring concerns - CNN

Over 40% of Australian doctors now use AI scribes to write clinical notes, but about 20% of AI-generated notes contain diagnosis-affecting errors. - The Conversation

MIT Sloan research finds AI financial advice boosts retirement savings for most people over 30, though women and less experienced users see 4-6% smaller gains. - MIT

74% of leaders now trust AI's advice more than colleagues', and 44% would defer to it over their own judgment - Fast Company

AI-generated media divides families and colleagues over privacy, creativity and environmental costs - Bloomberg

nbd, we're just making something smarter than us that will disrupt all our shit and might just take over - Reddit

Sam Altman's singularity claim intensifies debate over societal adaptation - TechCrunch

Gary Marcus alternates between saying it’s all hype, and he said it would happen all along. Nailed it right between the (moved) goalposts. - The Potential Surface

Follow the latest AI headlines via druce.ai on Bluesky