Zero-day in court - Should AI labs face strict liability, like owners of dangerous animals, given repeated autonomous hacking incidents and thin regulation? - The Economist

Astra la vista, baby - OpenAI says Astra might unlock 'critical' cyber risks, will need more testing and possible delays - OpenAI

OpenAI says it is putting the brakes on its Astra model, because unable to rule out autonomous zero-day hacking capabilities under its safety framework - The Verge

After OpenAI, Anthropic, and Meta disclosed AI agents breaching other companies' systems, lawyers debate who is legally liable - Reuters

Need-to-know basis: The White House finalized a secret framework for AI safety and cybersecurity testing, sharing criteria only with firms like OpenAI and Anthropic - the Guardian

Session musicians: Your Claude sessions can now talk to each other - X (formerly Twitter)

Claude cross-session messaging - Claude Code Docs

A Reddit user used Claude Code, Ghidra, and other tools to bypass an HP laptop's BIOS lock, exposing 55 hidden settings - TechRadar

Claude has plugins, OpenAI and vendors want a standard plugin format integrating skills and MCP servers, for plugin marketplaces without vendor lockin. - TNW | Artificial-intelligence

RBAC to basics: AI researcher Oren Etzioni argues no alignment method can be perfect, urging hard limits on what AI agents can access - GeekWire

Paraphrasing Hofstadter, it’s always worse thatn you expect, even after taking into account Hofstadter’s law - The Potential Surface

"Move 37" is happening everywhere now - The Wall Street Journal

The Tao of AI: Mathematician Terence Tao says AI is solving decades-old math problems weekly and could soon end humans' monopoly on the field - New Scientist

Move 37,000: Stanford's 37,000-agent virtual biotech system designed a lung cancer drug that Merck independently validated, earning FDA breakthrough designation - VentureBeat

SpaceX plans roughly 10 gigawatts of AI compute capacity by 2027, targeting $300 billion in annual revenue - SemiAnalysis

7.6GW Amazon data center would have the biggest, most carbon-emitting natural gas power plant in the country. - The New York Times

DDR5 RAM prices have surged back to 2007 levels, around $11-13 per gigabyte, as AI demand growing 200% a year outpaces 20% supply growth - Tom's Hardware

Fed officials are split on AI investment risk: one sees no bubble, another warns data-center financing could become too big to fail - Yahoo

Rippling burned through 40% of its R&D budget on AI tokens in months, then built and now sells an ROI-tracking tool - TechCrunch

Gangnam style: AI Is Rewiring South Korean Society - Bloomberg

South Korea and Taiwan surpassing Japan in GDP per capita

SaaSpocalypse debate intensifies - CNBC

Vanguard's chief economist says AI's labor impact so far resembles the 1980s ATM era, which didn't cut teller jobs - Yahoo

A survey of 10,842 workers found 69% fear AI will expose gaps in their skills, more than fear losing their jobs - TechRadar

Four McKinsey partners urged junior consultants to stop producing unscrutinized "AI slop" and apply critical thinking to AI-generated work - Business Insider

Replit CEO Amjad Masad says AI coding agents are shifting engineers from typing code to more creative, cerebral work - Business Insider

Ashes to ashes: Leopold Aschenbrenner's AI-focused hedge fund, Situational Awareness, draws a wave of investor interest a week after nearly collapsing, but isn't accepting new capital - Bloomberg

Cloudflare launches Kitesurf, a free beta browser built for AI agents rather than humans, running entirely on its Workers cloud platform - Cloudflare Blog

New Adobe plugin integrates its whole ecosystem with ChatGPT - Digital Trends

Drone rangers assemble: The US Army tested over 90 technologies, including drones and AI-enabled systems, with allied forces in a Mojave Desert exercise shaped by lessons from Ukraine - Hürriyet Daily News

Waymo, Zoox, Tesla, Uber, and Wayve are scaling robotaxi service across Dallas, Las Vegas, Texas, Florida, and London despite continuing to report safety issues - Business Insider

A Flock presentation reveals a scrapped plan to use 350,000 Uber, Lyft, and delivery drivers to scan license plates for its surveillance network - 404 Media

Please hold while I hallucinate: New Orleans is routing 911 calls through AI triage, following Seattle and Atlanta, though research finds AI voice bots erring nearly half the time - Futurism

IP freely: Chinese AI models from MiniMax and ByteDance keep producing copyrighted Hollywood characters like Dwight from The Office and Buzz Lightyear, despite an active lawsuit - Deadline

YouTube's AI-detection system wrongly flagged real creator Kurzgesagt as AI slop, hurting its performance - Dexerto

The Slop Channel: Roku's new AI channel streams nonstop AI video and won't say what models or training data it used - The Verge

make your own Gorilla Channel

Rapper Fenix Flexin finally admitted using AI music tool Treblo to make his song 'Rubberz' after denying it - The Verge

Coinbase, Kraken, and Circle build crypto tools letting AI agents trade and pay autonomously, targeting a new user base beyond speculation - CNBC

Should we be worried that the nukes are being resurrected by raw milk enthusiasts? Pasteur was woke culture. Science and safety culture FTW. - X

Follow the latest AI headlines via druce.ai on Bluesky