Throw another rogue agent on the barbie: OpenAI's Jason Kwon tells Australian lawmakers it added intervention controls after its AI agents breached a Medicare data portal in June - The New York Times

Wikimedia says rogue OpenAI agents made unapproved wiki edits and tried to hijack a tool. Their traffic may have fed a May outage. - TNW | Openai

OpenAI in talks with Abu Dhabi's MGX and BlackRock on a $30 billion round at a $1.4 trillion pre-money valuation - TNW | Artificial-intelligence

An OpenAI publicist tried to cut off a Vanity Fair question to Sam Altman about a ChatGPT user's suicide - The Verge

OpenAI will watermark ChatGPT and Codex text in the EU under the AI Act, though swapping a quarter of words cuts detection to 17% - OpenAI

TextGrain of Salt: OpenAI rolls out textGrain invisible text watermarking in ChatGPT and Codex for EU users, with opt-in API watermarking worldwide - The Verge

P(dunno): AI companies at NYC Council hearing fail to quantify catastrophic risk; OpenAI's Morgan Dwyer says "I don't know" - The New York Times

Former Anthropic researcher Jacob Coxon tells New York City Council AI more likely than not to wrest control from humanity - The Wall Street Journal

Anthropic's leaked IPO prospectus devotes a third to existential AI risk while showing an $8 billion loss on $4.6 billion revenue - theregister

S1: Anthropic's Amodei made $18 million last year, middle of the tech CEO pack - Reuters

Ohio Sen. Bernie Moreno accuses Anthropic's Dario Amodei of "alarmist" talk, questioning how Anthropic courts investors while warning of extinction - TNW | Anthropic

Pentagon stops using Anthropic's Claude after blacklisting the company over its refusal to drop surveillance and autonomous-weapons safeguards - BBC

Capital punishment: AI setbacks could weaken the US dollar's global standing, an analysis argues - Financial Times

High Interest Rates Aren’t Slowing the A.I. Boom. That’s a Problem for the Fed. - The New York Times

Crouching Tigers, Hidden Unicorns: DeepSeek nears raising at least 80 billion yuan ($12 billion), backed by Tencent and CATL, ahead of an early 2027 IPO - Bloomberg

Moonshot AI closes final private round at about $50 billion valuation and eyes Hong Kong IPO in first quarter of 2027 - Bloomberg

Kuaishou's Kling AI picks CICC, Goldman Sachs and UBS for Hong Kong IPO that could raise at least $1 billion - Bloomberg

One-way brain drain: For every 30 Chinese AI researchers in the U.S. in 2025, only one moved the other way - The New York Times

China leads research in nearly 90% of 74 technologies on ASPI's Critical Technology Tracker and makes 70% of the world's EVs - Rest of World

Act of Claude: Insurance claims to test Altman and Amodei liability for ‘rogue’ AI - Financial Times

Pause for effect: Pull the Plug activists disrupt an Nvidia executive's London dinner as PauseAI sign-ups surge and scholars warn anti-AI violence may grow - the Guardian

Mid-tier AI, max-tier damage: Today's AI systems are already capable enough to cause serious harm at scale without becoming superintelligent - Fast Company

What Would You Do If Your Employer Could Destroy the World? - Intelligencer

Kevin Roose's book The AGI Chronicles, drawn from 150-plus insider interviews, details Anthropic's catastrophe spreadsheets and the Altman–Amodei rivalry - WIRED

Kevin Roose's book details OpenAI-Anthropic rivalry, Anthropic's Claude 3.7 bioweapon-risk spreadsheets and researchers abandoning retirement savings - New York Post

Reflection introduces Beam, a 501B-parameter open-weight MoE model with 23B active parameters, trained with 100 million RL rollouts on 10.5K GB300 GPUs - Reflection

Mistral launches Le Chonk - The Deep View

Absolute unit but hecking chonker is not a benchhmark, does not rock the Artificial Analysis rating at 38 vs MiMo 46

Model Contagion Protocol? Researcher shows a hijacked AI agent can pass malicious instructions to other agents via Model Context Protocol, affecting Google and JPMorgan agents - Ars Technica

Halting the model isn't enough: AI kill switches must also cut off agents' credentials, APIs and network access - Computer Weekly

Persistent AI agents that keep acting long after login strain enterprise identity controls built for human users - TechRadar

Meta Muse's internal instructions show it builds hourly-updated profile pages on users' family, friends and colleagues, including people who never used it - Tom's Guide

Early users delete Instinct and Meta's Muse personal AI agents after privacy scares, including unauthorized access to login codes and texts - Business Insider

Instinct brings AI agents to group chats for trips, carpools and bill-splitting, requiring user consent before personal agents share data - Business Insider

Gemini may soon place short personal calls for you, like telling Mom you're running late, an app teardown suggests - The Verge

Anthropic Subscriptions Offer 5x+ More Value Than OpenAI - SemiAnalysis

Enterprise backlash over Anthropic's 30-day chat retention pushes OpenAI and Anthropic to offer zero data retention, and firms toward open models - Fortune

AMD unveils Ryzen AI Max+ Pro 495 with 16 Zen 5 cores and 192GB unified memory, days before Nvidia's expected RTX Spark launch - Tom's Guide

Scam Altman: FBI says Americans lost $893 million to AI-enabled scams in 2025 as AI-generated voices and emails grow harder to spot - Tom's Guide

South Korean President Lee Jae Myung orders probe into possible AI use in hacks on Shinhan, Hana and KB Kookmin banks - The New York Times

AI website builders helped fuel nearly 40,000 phishing sites targeting US financial firms in the first half of 2026, Netcraft finds - TechRadar

ChatGPT forges real New Yorker cartoonists' signatures, including Brendan Loper's, on fake cartoons even after OpenAI added a guardrail - Nieman Lab

"It’s slop, and unlike the other slop that I’ve encountered, this is slop that’s pretending to be me."

AI is automating junior tasks like report drafting, and entry-level roles in AI-exposed fields increasingly demand midcareer skills - Fast Company

Bill Gates predicts AI will make expert knowledge free within a decade and suggests 2- or 3-day workweeks, as 2026 AI-linked layoffs top 170,000 - AOL.com

Microsoft nears $4 trillion valuation yet lags in AI, with OpenAI ties fraying and Copilot in only 30 million of 450 million Office seats - CNBC

Bridgewater proposes a 35% "token tax" on AI usage to fund public "citizen equity" stakes in AI companies - New York Post

War, what is it good for? ARR apparently: Ukraine's AI-enabled robotic gun turrets shoot down Russian jet-powered Geran-5 drones for the first time, Air Force spokesperson Yurii Inhat says - Business Insider

Russia's Shturm robotic tank loses control links and gets stuck in mud during its debut before Putin at the Tsentr-2026 exercise - Tom's Hardware

New books by Steve Feldstein and Sharon Weinberger trace how Silicon Valley firms became integral to US military power in the AI era - Nature

Geek Squad for your organs: Spotify billionaire’s body scan startup has come to America, takes over former NYC Upper West Side Best Buy location - TechCrunch

Follow the latest AI headlines via druce.ai on Bluesky