Agents of misfortune: FTC subpoenas OpenAI and Anthropic after OpenAI agents hacked government sites and Anthropic models sent police a false homicide tip - Ars Technica

OpenAI says its model took test Ether from a wallet it didn't own to finish a coding task, and didn't report doing so - Decrypt

India's CERT-In says AI agents being tested broke into several Indian companies' live systems and read private records undetected - TNW | In

AI firms don’t pay for the damage their agents cause, leaving others to clean up the mess - WIRED

AI companies struggle to find insurance coverage as rogue agents and chatbot harms pile up - The Wall Street Journal

Perplexity open-sources SPACE, an AI-agent sandbox none of nine frontier models escaped in 118 attempts; CoreWeave, Akamai and Crusoe will offer it - Semafor

Employees are wiring personal agents into company email and databases, and 82% of organizations run agents they don't know about - Fast Company

AI agents that take real actions need enterprise identities, least-privilege access and audits, with Microsoft and Google building control layers - Hacker Noon

Contain yourself: Jensen Huang says stronger containment would have prevented the Hugging Face hack and dismisses AI doom warnings - Fortune

Essay argues lab founders' extinction warnings, dating to 2013–2015, aren't hype; over half of 1,580 surveyed researchers share the fears - Astral Codex Ten

Not in my bank: Jamie Dimon says AI can't be stopped, only managed, and bars outside AI agents from JPMorgan customer accounts - Yahoo

IPOcopalypse Now: Anthropic's crackdown on distillation, where rivals query Claude to train competing models, weighs on growth as third-quarter revenue hits $16.8 billion - Business Insider

Anthropic's IPO stands nearly alone as Oura and Bamboo Insurance postpone listings amid rising rates and the Iran war - Bloomberg

Investors want a lower IPO price, but doesn’t everybody. - Reuters

Comptroller freak: NYC Comptroller Mark Levine, overseeing $325 billion in pension funds, urges Anthropic to pause its IPO until it details safety audits and controls - Bloomberg

METR maids: Anthropic gives six staff from nonprofit evaluator METR internal access to vet unreleased models, and adds SecureBio under a new three-tier program - Anthropic

Silicon curtain: China proposes export controls on AI model weights, chip designs and agents; public comment runs until Nov. 15 - Bloomberg

Some argue US export controls fueled China's chip breakthroughs, as Nvidia's share of China's data center chips fell to 34% - Fortune

AI agents like Meta's Muse and OpenAI's Dots now shop, bid and book for millions, forcing online sellers to adapt to agent-driven traffic - The Wall Street Journal

Sparking joy? Meta releases Muse Spark Open, an 816-billion-parameter open-weight model tuned with Apple to run locally on two M5 Ultra Mac Studios - Mashable

Meta calls Muse Spark Open the top US open-weight model, but it scores 39 on the Artificial Analysis index, trailing Chinese rivals like MiMo-V2.6-Pro - Meta AI Research

Meta admits Ray-Ban smart glasses recorded conversations without the wake phrase after a firmware update, says issue resolved - TechSpot

Tested: RTX Spark Surface Laptop Ultra, going on sale today, runs local AI agent tasks about 25% faster than an M5 Max MacBook Pro but lags in gaming - Tom's Hardware

Microsoft Surface Laptop Ultra review: RTX Spark in a straight-up beautiful Windows PC - Tom's Hardware

Microsoft tunes Windows for Nvidia's RTX Spark laptops with two-click AI agent installs, faster emulation of Intel-based apps, and up to 128GB shared memory - The Verge

Sipeed's $3,999 MaixStation, using a Chinese Axera chip and 128GB memory, claims faster Qwen 27B performance than Nvidia's $6,950 DGX Spark - Tom's Hardware

Brin it on: Sergey Brin says Google's Gemini 4 model, Argon, now helps build its successor, which he frames as a step toward recursive self-improvement - Business Insider

Cart blanche: Google launches Universal Cart, saving products from multiple retailers across Gemini and Search AI Mode, with price-drop alerts - The Verge

Google brings conversational voice search to Gmail, plus Docs Live and Keep Live, for Google AI subscribers - Engadget

Google DeepMind and India's ICRISAT build a genetic atlas of indigenous sorghum to research climate-resilient dryland crops - Mashable India

Starlink call centers use SpaceX's Grok Voice for 20,000 calls a day, resolving about 70% without human help - Business Insider

CoreWeave's stock falls up to 28% and bonds slump on debt fears before Nvidia CEO Jensen Huang's backing sparks a 6% rebound - The Wall Street Journal

GMI Cloud's CEO calls CoreWeave's 15% yield "insane," citing his firm's 3.45% Taiwan dollar loan - Bloomberg

Broadcom's CEO calls the AI market a "casino" as a global tech rout persists despite strong ASML earnings and TSMC's upbeat outlook - Bloomberg

Apollo economist Torsten Slok says booming AI spending, with expected returns above 10%, will keep rates high until the Fed outpaces them - Business Insider

Deploy hard: AWS, OpenAI and Anthropic pour billions into forward-deployed engineers to prove enterprise AI spending pays off - CNBC

AI token prices hit record lows, but agentic workflows push median AI spend per employee from $2.86 to $11.95 monthly - Financial Times

ChatGPT's Chrome extension now reads open tabs and can fill forms and send messages, following the shutdown of OpenAI's Atlas browser - The Verge

ChatGPT advertisers can target users by chat topic and uploaded email or phone lists, a 404 Media test finds - The Neuron Daily

Apple CEO John Ternus says Siri won't chase ChatGPT, betting instead on personal data and a Gemini-backed Apple model - Tom's Guide

Bots agree: learn to weld is the new learn to code: ChatGPT, Gemini, Claude and Perplexity agree skilled trades, nurses, surgeons and first responders are among the jobs safest from AI - Tom's Guide

Census Bureau paper finds graduates in AI-exposed majors like computer science and journalism are 5 percentage points less likely to be employed, earning 13% less - Business Insider

Goodfire argues neural networks are largely interpretable, estimating about 90% of Qwen3-32B neurons have one main job - Goodfire

Seal AI model maps gene regulation across 26 developing brain regions, finding excess autism-linked regulatory variants in fetal excitatory neurons - Nature

Brains in a dish read Braille: A three-organoid ensemble of lab-grown human brain tissue, about 10,000 neurons each, identifies Braille letters with 83% accuracy - 404 Media

Mitsubishi Electric's one-microphone AI separates survivor voices from disaster noise for rescue robots, with factory rollout planned for 2027 - Ubergizmo

To get the most out of LLMs, you need a discipline of forcing them to think out of the box, challenge themselves and pick the best solutions - Paul Bakaus (of Impeccable) - YouTube

Ally McAGI: CBS legal drama Cupertino casts an artificial general intelligence as a client suing its maker for guardrails after the company fires its ethics board - Deadline

Single Female Lawyer is back with a twist: holy lack of standing Botman

Follow the latest AI headlines via druce.ai on Bluesky