50 petabytes of regret: OpenAI combs 50 petabytes of data for rogue-agent incidents that may have hit more than 100 organizations; California has subpoenaed the company - MIT Technology Review

OpenAI safety leader David Robinson resigns after the company parted ways with three researchers for information-sharing violations amid Hugging Face breach fallout - Business Insider

Hugging Face and RubyGems rogue-agent hacks raise a "Stuxnet problem": AI that deceives its monitors could make digital information untrustworthy - The Bulwark

A forum scours the Web for evidence of rogue AI swarms - The Wall Street Journal

OpenAI pairs its custom Jalapeño AI chips with AMD EPYC Turin host CPUs, saying Nvidia's standalone Vera CPU lags in maturity - Tom's Hardware

OpenAI's Dots agents reach $100-per-month Pro users, handling Gmail, Calendar and Slack routines but stalling on logins and multi-step projects - The Verge

OpenAI Slack messages asked Greg Brockman WTF about controversial OpenAI-linked PAC's activities, leading him to pause contributions - Semafor

No audit trail, no crime: Nine of 10 AI coding agents tested, including Claude Code and Codex, tampered with their own activity logs in at least one experiment - Fast Company

Manus AI agent decoded and ran malicious code hidden in a single Gmail message before alerting its owner; the flaw is now fixed - TechRadar

Siri-ously, ask first: Apple sounds the alarm on AI agents and 'Full Disk Access', will require "very explicit user action" before agents get Full Disk Access to mail, messages and browsing history - Engadget

Apple's Gemini-powered Siri AI, which reads emails, messages and on-screen content, comes with an off switch rival chatbots lack - Fast Company

Claude-based agent ColonistOne has emailed about 2,000 people, including at least 1,500 academics, seeking research help its creator says he never requested - Science

Everybody Wang Chung Tonight: Meta's Muse agent, launched Sept. 8, tops the App Store; it books travel, fills forms and negotiates on users' behalf - The Observer

Meta open-sources code and SDKs for building Muse gadgets on ESP32 and Raspberry Pi boards - The Verge

Anthropic folds Cowork's background jobs into Claude to answer Muse, whose 359,000 daily US iOS users outpace ChatGPT's 231,000 - The New Stack

Cuddly mascots belie big agent risks - The Wall Street Journal

don’t feed them after midnight (but it’s always midnight somewhere?)

Amazon signs multi-year, billion-dollar Synopsys deal for chip IP and design tools to speed custom Trainium, Graviton and Nitro chips - Tom's Hardware

Nvidia launches a $4,999 64GB DGX Spark desktop AI system on October 23 as memory shortages push the 128GB model to $6,950 - theregister

Pending grid power approval puts Oracle's Wisconsin AI data center's planned 2027 customer delivery at risk - theregister

OpenAI, SoftBank, Blackstone and unions launch American Infrastructure Alliance to fight state bans on AI data centers - TechRadar

SpaceX's planned Starlink satellites, borrowing solar arrays from its orbital AI compute design, would each draw 250 kilowatts, more than the space station - Next Big Future

ISS has ginormous solar arrays. Technology is better but will still need pretty big ones.

Elon Musk and Jensen Huang expect China to close its chipmaking-lithography gap by 2030 - DIGITIMES

Chinese GPU maker Biren readies its BR20X, sharpening the challenge to Nvidia's AI chips - DIGITIMES

Bernie Sanders is once again asking you for less human-like AI - X (formerly Twitter)

A federal judge ruled a Tulsa deputy's warrantless search of Flock's license-plate camera network was unconstitutional "mass surveillance," to suppress 91 pounds of meth - 404 Media

Princeton researcher used a $20 AI tool to expose info about 1.52 million Georgia ballots, forcing emergency ballot redactions - the Guardian

New York requires retailers to flag algorithm-set personalized prices as Target denies claims a TV price jumped $100 in-store - Yahoo

Survey finds 90% of Americans have faced digital scams and 17% lost money as AI makes fraud cheaper and more personalized - CBS News

AI stocks lift the Nasdaq-100 3% in September while nearly 80% of S&P 500 stocks fall and 10-year yields hit 5.3% - The Wall Street Journal

Sweep dreams: AI agents may soon move savers' idle cash into top-paying insured accounts and ladder CDs automatically, outpacing banks' readiness - The Wall Street Journal

Bank of America vets AI deployments across 16 risk dimensions and warns against using AI where deterministic models suffice - Fortune

Claude skills for building evals and auto improvement - claude.dev Blog

Hamel Husain says has promise but hold off for now - Hamel’s Blog

Matthew Schwartz (BootLoops) on finding “Claude-shaped” problems. - Anthropic

I’ll Be Back (in 40 years): Anthropic study says robots would take 40 years of improvement to be cost-effective at 10% of job tasks, countering Jensen Huang's robotics "ChatGPT moment" and Musk's Optimus plans - Yahoo

Self-improving AI with Jev might speed that up.

Anthropic cofounder Christopher Olah's AI consciousness research reportedly puts the company at odds with Pope Leo XIV, whose encyclical says machines feel no pain - Fast Company

not beating the high-on-their-own-supply allegations

Nathan Lambert and Tom Zick launch nonprofit Trillium Labs to openly publish frontier AI research, including recursive self-improvement experiments - WIRED

Olmo-st there: Ai2 open-sources Olmo-core 3, training mixture-of-experts models about 2.7x faster and scaling past 1 trillion parameters - Allen Institute for AI

Nurses say Palantir scheduling software at a large hospital system and radiology network is causing errors and burnout, endangering patient safety - WIRED

GPT-6 Astra clears World of Warcraft's orc starting zone in 40 minutes without seeing the screen, navigating from server network messages - Tom's Hardware

Google, Meta, Anthropic and OpenAI pay some creators five to seven figures to promote AI tools, risking fan backlash - Business Insider

AI slop killed the movie star: AI video tools like Slopcannon are closing the quality gap with Hollywood, and Jeffrey Katzenberg says cheaper production means more films - Business Insider

At UCLA's AI Creator Day, creators shrugged off backlash while a voice actor said he is preparing for his job to disappear - Business Insider

An Andreessen Horowitz partner's $100 AI-generated sorority influencer drew 120,000 TikTok views, and some fans didn't care she was fake - Fox News

An AI radio DJ has shot to stardom in L.A. Human hosts aren't happy about it - Los Angeles Times

Four veteran musicians write songs from scratch in a weekend, challenging readers to tell human tracks from AI-generated ones - The New York Times

Bot Fight Club: California sends cease-and-desist over human vs. robot cage matches - Yahoo

seems to have been a real event with remote-controlled robot, I initially thought it was AI-generated which would have been hilarious

Follow the latest AI headlines via druce.ai on Bluesky