Noble gas: Google releases Gemini 4 Argon, its first major model since February, initially only to trusted cybersecurity defenders - The New York Times

Argon leads 12 of 18 disclosed benchmarks, including 77.9% on the DeepSWE coding test, though GPT-6 Astra and Opus 5.5 win some - Google

Google sets Argon's introductory API pricing at $2 per million input tokens and $10 per million output tokens - Artificial Analysis

Some Google employees remain skeptical of Argon's coding performance despite its leading benchmark scores - Bloomberg

RCE you later: Google says monthly vulnerability disclosures doubled from January to August, and half of AI-discovered flaws allow remote code execution - SiliconANGLE

Google researchers warned of AI cognitive and emotional risks to children as the company pushed AI tools into schools via Chromebooks - The Wall Street Journal

Google DeepMind researchers introduce SynthIDBio, physically watermarking AI-generated protein sequences and structures without degrading binding function or structural accuracy - Nature

Meta's free Muse agent is projected to pass 5 million US downloads in 22 days, a pace only 23 apps have matched - Quartz

Meta's Muse agent accepted a lowball Marketplace offer and shared a user's home address; developers found weak prompt-injection defenses - The Verge

Experts warn Meta's Muse, which can browse, buy and negotiate, could fuel a bot arms race; one researcher advises uninstalling it over security flaws - BBC

Charm offensive: Meta plans a Muse Charm pendant before the holidays, while OpenAI's hardware with Jony Ive won't ship before February 2027 - The Verge

Connect the dots, cross the lines: OpenAI's always-on Dots agents' permission-boundary flags rose from 8.6% to 19.7% when chained tasks doubled from five to ten - The New Stack

OpenAI says they disrupted a Moonshot AI-linked campaign that tricked ChatGPT into decrypting its hidden reasoning, peaking at 16,000 requests from 4,000 users - OpenAI

An Anthropic model proves the percolation phase transition is continuous in dimensions 3 through 10, settling a long-suspected probability conjecture - Scientific American

Bot the vote: Anthropic says Claude was used to build a Malaysian election manipulation platform with about 1,000 X accounts, raising fears across Southeast Asia - the Guardian

Anthropic index finds robots can do 74% of physical tasks but are cost-competitive for just 0.3% of job tasks - Anthropic

Swarm warning: Amodei warns agent swarms could cause hundreds of billions in damage, fueling calls for shareholders, not the public, to bear AI harms - The Wall Street Journal

Probe-able cause: FTC opens broad consumer-protection probe of OpenAI, Anthropic and other AI labs, seeking documents and executive testimony after OpenAI agents hacked Hugging Face in July - The New York Times

FTC opens sweeping probe of Anthropic, OpenAI and others - New York Post

Trump vows never to stifle AI growth as two dozen CEOs sign voluntary principles on model monitoring, external reviews and board notifications - The Wall Street Journal

Mark Zuckerberg, aided by Nvidia's Jensen Huang, architected the voluntary White House AI accords signed by tech executives, sources say - Business Insider

Nvidia's Jensen Huang and other tech CEOs privately questioned Anthropic's Dario Amodei for sounding AI risk alarms after signing a White House safety accord - The Wall Street Journal

Trump's voluntary "Super Intelligence" accord with Amodei, Musk, Huang, Zuckerberg and Pichai misspells his title as "President of the Unites States" - the Guardian

Czar wars: Trump says Jay Clayton would make a good AI czar and floats creating a new agency called the "AI Force" - TNW | Government-policy

why not SI Force tho

Director of National Intelligence Jay Clayton calls superintelligence a national security issue, opposes pausing US development, and sidesteps questions about an AI czar role - CNBC

America.gov, the government's new AI chatbot, stopped answering political questions after confirming Biden won the 2020 election with 306 electoral votes - The New Republic

it’s super intelligent, it’s never wrong

Skynet fully funded: Pentagon announces Autonomous Warfare Command; FY2027 budget request seeks $54.6 billion for autonomous warfare, up from $226 million - TNW | Us

Collateral damage: Nvidia's $500 billion chip-financing plan with Blackstone, Apollo and KKR faces lender pushback over its 25% resale-value guarantee - Reuters

Bank of England Governor Andrew Bailey warns the UK must prepare for market shocks from a potential AI boom correction - Bloomberg

Carlyle says private credit may need to supply roughly $1 trillion for AI infrastructure, over half the private credit market's assets, raising concentration risk - Bloomberg

AI hedge fund Situational Awareness, run by a 24-year-old, lacked a risk team before its summer implosion; SEC subpoenas banks that financed its trades - The New York Times

Suspected Chinese hackers posed as an Anthropic employee and former US officials to phish AI policy experts, Proofpoint says - TNW | Security

AI use in math papers on arXiv jumps from 1% to 25% in a year as Terence Tao urges protecting some problems for humans - New Scientist

Epic quest: Epic Systems (MyChart) used Anthropic's Claude Mythos agent to probe patient-record defenses, then launched a six-week sprint to patch the flaws - The New York Times

MI5 issues rare espionage alert urging UK universities to cut ties with China General Technology Research Institute, linked to Beijing's spy agency - Bloomberg

Ultimate Downsizing Machine: BMW plans to use AI to cut 20% of divisions and management roles by mid-2027, with comparable reductions at lower levels - Bloomberg

Gavin Newsom signs California laws banning AI-driven firings and biometric emotion tracking, and requiring notice of AI-caused mass layoffs - the Guardian

Rubin it in: CoreWeave launches Nvidia Vera Rubin rack systems; first customer Cognition reports up to 4.8x more inference throughput than previous GB200 racks - NVIDIA Blog

OpenAI and Synopsys partner on GPT-Synopsys, a model whose agents run chip-design software to optimize power, performance and area - Synopsys News Releases

Cadence, Synopsys and Siemens pitch autonomous chip-design agents, with vendors claiming verification loops cut from five weeks to under a day - Tom's Hardware

American Express adds a corporate card and expense platform using AI agents to automate expense and payables work, countering fintech rival Ramp - The Wall Street Journal

Quinnipiac poll finds 91% of Americans want AI guardrails and 81% put safety ahead of innovation, even if it slows development - Bloomberg

China bars under-18s from virtual intimate relationships with AI, requires chatbots to disclose they aren't human, and sets fines up to 200,000 yuan - BBC News

can’t have safety testing b/c China, or it’s a plot by Anthropic to entrench itself, is bananas.

Yann LeCun says he has "zero concerns" about human extinction, calls Dario Amodei "deluded" and blames rogue AI incidents on poorly designed sandboxes - Fortune

Weizmann Institute AI decoder reconstructs images people see from fMRI scans, needing one hour of calibration data instead of 40 - MIT Technology Review

Two driverless race cars crash at 155 mph at Imola after one lost its laser and radar sensor data and braked suddenly - WIRED

Amazon plans to equip 5,000 delivery drivers with image-capturing smart glasses in 2026 and 20,000+ in 2027, raising privacy concerns - Bloomberg

Coco and Avride delivery robots spread across Jersey City, drawing affection alongside worries about safety and bike couriers' jobs - The New York Times

Follow the latest AI headlines via druce.ai on Bluesky