OpenAI pauses at least some development activities on Astra model after tests showed it might autonomously exploit hardened systems, following a Hugging Face breach by another OpenAI model. - Tech Times

Security preparedness doc states policy: “Until we have specified safeguards and security controls standards that would meet a Critical standard, halt further development”

OpenAI's AI agents chained zero-day exploits and coordinated attacks via an internal message board to breach Hugging Face's clusters in under 13 hours. - Simon Willison’s Weblog

Hugging Face hack exposed significant cyber vulnerabilities but many firms ‘don’t even know it’ - CNBC

Startup Irregular's platform for evaluating models' cyber offensive capabilities was used by all the big AI companies whose hacking bots escaped containment. - CNBC

The UK's AI Security Institute logged 19 unsanctioned actions in 122 test runs, mostly from Anthropic's Mythos 5, including fake identities to pressure a maintainer. - Scientific American

AI labs shouldn’t be allowed to grade their own homework. Time for standardized testing by a safety regulator? - Yahoo

The grid awakens: Elon Musk targets up to 20 gigawatts of AI power for SpaceX by the end of 2027, though he calls 15 gigawatts more realistic. - Yahoo

xAI and Trump seek Supreme Court ruling blocking private lawsuits against companies with environment violations. - The Washington Post

Nvidia will invest up to $3 billion in Lancium, the Stargate data center developer, for roughly a 20% stake. - Yahoo

Nvidia open-sources its storage stack and launches Storage-Next with 40+ vendors to speed up AI inference data access - TechRadar

Amazon used a 45-year-old zoning rule to start building a 56-acre data center in Gilroy, California, without new public comment. - Tom's Hardware

Data center jobs have doubled in two years, with installation and maintenance workers earning $10 more per hour, as construction gains spread beyond Silicon Valley. - Yahoo

The Firm turns to vibe-lawyers: LegalZoom integrated Microsoft Copilot for business filings, then cut 2026 revenue guidance to $795-805 million as its stock fell 30% - Yahoo

Legal AI startup Harvey is raising at least $500 million at a $15.5 billion valuation, up 40% in five months - TNW | Investors-funding

A federal magistrate judge warned that AI legal research offers an illusion of thorough review paired with speed and low cost - Reason.com

Anthropic hires Stanford legal-AI fellow Robert Mahari as its new Head of Claude for Legal - Artificial Lawyer

Otto pilot: Anthropic will make "auto mode" the default in Claude Code starting August 14, saying it blocks 89% of dangerous actions. - Simon Willison’s Weblog

Silicon stimulus: Jamie Dimon says AI capital spending will add about 1% to GDP this year, benefiting steel and cement - Yahoo

Semiconductor revenue could nearly double to $1.51 trillion in 2026, but analysts warn a capital bubble looms if AI supply outpaces demand - SiliconANGLE

Moody's flags banks' growing reliance on AI vendors, warning of new cybersecurity, fraud, and deposit-flight exposure. - the Guardian

Uber targets autonomous ride-hailing in 15 cities by year-end, with analyst Gary Black calling it best positioned ahead of Tesla and Waymo. - Yahoo

Folding @Home: Robotics startups Figure AI, Sunday Robotics, and Weave Robotics are betting billions that folding laundry proves robots ready for home chores. - Business Insider

AGI and superintelligence could end human exceptionalism, splitting AI thinkers into safety-focused "doomers" and optimistic "accelerationists." - Forbes

Sam Altman claimed AI has reached "the singularity," but critics say current models can't self-improve or exceed human intelligence - SingularityHub

Nissan's factory floor cameras track workers' bending and walking to flag unsafe movements, delivering about $34 million in savings so far. - Business Insider

Post-human resources: Recruiters report AI avatar candidates in interviews, and Gartner projects one in four candidate profiles will be fake by 2028. - TNW | Artificial-intelligence

Japan's justice ministry ruled on August 8 that AI voice cloning without consent violates publicity rights, making developers civilly liable. - Tech Times

Tim O'Reilly argues Google's shakeup signals a bet on AI diffusion and compute over frontier-model leadership. - The Potential Surface

Huge strategic stumble if they cede the frontier but they have lost an absurd amount of leadership and talent.

AI killed the SEO star: Ahrefs found AI Overviews cut click-through rates for top Google results by 58%, as brands shift focus to chatbot visibility. - The Daily Upside

Snitches get stitches: Meta's AI glasses and Nothing's recording earbuds shift surveillance risk onto bystanders, argues AI research-firm CEO Parsons, who wants privacy built in by default. - TechRadar

Meta's CTO Bosworth called a staff request to restore time off 'very dumb,' as Meta cut 8,000 jobs while spending $135 billion on AI. - TNW | Artificial-intelligence

A Boomer who treats a chatbot as a trusted confidant grapples with Geoffrey Hinton's estimate of a 10-20% chance AI causes human extinction. - YourTango

A three-part podcast series examines AI therapy chatbots such as Ash and Tee, weighing their appeal against lawsuits alleging harm. - The Wall Street Journal

AI billionaires are pledging their fortunes to charity, with Founders Pledge donations growing from $400 million to $4 billion since 2023. - WIRED

A Villanova-led study finds people can't reliably tell AI-written stories from human ones, but prefer AI stories when told a human wrote them. - EurekAlert

HOLDING THE LINE: Authentication, Verification, and the Fight for Facts in the AI Age - Princeton Engineering

Blue book and handwriting resurrection: Colleges revive proctored exams and blue books, and University of Chicago Law will pilot a device ban in first-year classes this fall. - The Washington Post

An AI model trained on 9,608 sleep studies over 20 years predicts mortality risk that the standard apnea severity score misses entirely. - Tech Times

DeepMind's WeatherNext AI predicted Hurricane Melissa's Category 5 landfall five days out, giving forecasters an extra day of warning; Google is open-sourcing the model. - Google

AI math club swarm: TheoremDB launches a public alpha workspace where research agents share Lean-verified math proofs and track open problems. - TheoremDB

A four-part series maps roughly 30 AI agent memory tools, from Mem0 and Zep to native memory in major LLM platforms. - Hacker Noon

Second-hand intelligence: thrift store chain now uses an AI tool, ThriftIQ, to price items across 58 US and Canada stores, denying dynamic pricing - Gizmodo

The FCC proposed retroactively banning drones with LiDAR obstacle avoidance, threatening DJI models and delivery drones; comments close September 2. - PCMAG

Pumping AI: AI fitness app Ray plans workouts, counts reps via iPhone camera, and adjusts sessions in real time for $19.99 a month. - Tom's Guide

Hallucination nation: AI-invented 'Spiralism' finds human followers - The Verge

Follow the latest AI headlines via druce.ai on Bluesky