OpenAI launched ChatGPT Work, a GPT-5.6 cloud agent that autonomously handles email, Slack, calendar, and GitHub tasks for Pro and Enterprise users first. - VentureBeat

OpenAI's GPT-5.6 Sol model proved a 50-year-old graph theory conjecture, showing any qualifying graph needs no more than eight loops to cover. - Scientific American

OpenAI launched Codex Micro, a $230 keypad built with Work Louder, featuring a joystick and live status lights for coding-agent tasks. - The Verge

OpenAI's new GPT-Red red-teaming system found real prompt-injection bugs and made GPT-5.6 six times more resistant to attacks before release - OpenAI

Anthropic is scheduling investor meetings for a possible mega-IPO, with a listing considered as soon as October. - Bloomberg

Blackstone, Hellman & Friedman, and Goldman Sachs backed Anthropic's launch of enterprise AI implementation venture Ode, a $1.5 billion effort with 100 engineers. - TechCrunch

Anthropic pushes states toward tougher AI safety laws while OpenAI urges Congress to codify state rules into one national framework instead - Business Insider

OpenAI's Chris Lehane says state laws in California, New York, and Illinois are building a de facto national AI safety standard. - OpenAI

Apple confirms it is building AI features with Google's Gemini and Nvidia models, a notable shift from its past rivalry - Fast Company

Pulls off Tim Cook face mask, reveals Jensen: Apple is shopping for chip companies and accelerating its own AI server roadmap after its chips proved inadequate to run Siri in the cloud, forcing it onto Nvidia in Google's cloud. - TNW | Apple

Chinese startup Moonshot's upcoming Kimi K3 model is expected to surpass Anthropic's Claude Opus 4.8, narrowing the US-China AI gap - Financial Times

Ctrl-CCP: Palantir CTO Shyam Sankar says Chinese open-source AI models are economic threats built on stolen US intellectual property - Bloomberg

Inkling and a prayer: Thinking Machines, founded by former OpenAI CTO Mira Murati, released a 975-billion-parameter open-weights model to rival leading Chinese open models - theregister

Mira Murati's Thinking Machines released Inkling, a 975-billion-parameter open-weight model with a 1-million-token context window that the company admits isn't the best available. - Thinking Machines Lab

Tests like deepseek-v4-flash here, best US open-source model, good for agentic tasks - Artificial Analysis

Clippy on the warpath: Microsoft is training salespeople to attack Anthropic and OpenAI on cost, security, and product completeness ahead of its new fiscal year. - Bloomberg

Musk rocketship falls from orbit: SpaceX shares fall below $135 for the first time since its June debut, erasing roughly $1 trillion in value - Financial Times

Musk's xAI, renamed SpaceXAI after a merger with SpaceX, battles internal chaos as new president Michael Nicolls sets a goal to match Claude's performance - Bloomberg

Oops Elon stole your secrets: xAI open-sources coding tool to regain trust after massive uproar because it uploaded complete directories including API keys. - Simon Willison’s Weblog

xAI sued a man accused of using Grok to generate child sexual abuse material, seeking damages and a ban on his access - The Verge

He was writing prompts, Grok was the one generating the CSAM tho?

Desert superbloom: TSMC pledges $100 billion in additional US investment, bringing its total Arizona commitment to $265 billion, while raising 2026 capex guidance to $60-64 billion - The Wall Street Journal

Unstoppable AI force meets immovable grid object: PJM's power auction hit its price cap, adding $6.3 billion to bills as AI-driven demand leaves the grid 6.8 GW short of its reliability target. - Fast Company

Bankers are marketing majority stakes in several data-center operators worth tens of billions as vacancy falls below 1% in markets like Northern Virginia. - BlockSpace Media

Hyundai to buy SoftBank's remaining roughly 10% stake in Boston Dynamics, becoming full owner, and deploy Atlas robots at its Georgia, US, plant from 2028 - Reuters

Physical AI, emotional damage: Hyundai's South Korea plant saw a partial strike over wages, AI, and deploying Boston Dynamics' humanoid robots onto the assembly line - The Wall Street Journal

Nvidia introduced Jetson T3000 and T2000 edge-AI computers, plus a 4-billion-parameter robot foundation model, targeting mass-market robotics deployment - NVIDIA Blog

Japan will buy 27,500 next-generation Nvidia chips to build a 140-megawatt national AI factory for robots, led by the SoftBank-backed Noetra consortium - TNW | Nvidia

Fujitsu is leading a Japanese robotics push using Nvidia technology to build physical AI that lets machines interact with the real world - The Washington Post

China is deploying humanoid robots faster than the US, sending them into factories and warehouses to teach robot software new skills - Bloomberg

Walden Robotics launched with $300 million and legless, wheeled humanoid robots already working eight-hour shifts at a Toyota plant - TNW | Artificial-Intelligence

Ukraine's $500 AI-guided FPV drone struck a Russian target at a record 68 miles — six times the previous range - TechRadar

Singularity of violence: CIA Director John Ratcliffe says Russian troops in Ukraine now survive only 20 to 30 minutes under AI drone attack - Bloomberg

Terminator interns: Foundation Future Industries expects to test weaponized humanoid robots as early as next year, following pilot programs in Ukraine - euronews

The Dutch navy is testing uncrewed vessels, drones and mine-mapping robots, aiming to surround crewed ships with autonomous systems within a decade - the Guardian

DeepMind researcher Alex Turner resigned in June after Google's Pentagon deal permitted classified military use of its AI, joining roughly 600 employee objectors - Business Insider

People’s republic of Nvidia: If you're not a billionaire in San Francisco, you're poor. - Business Insider

Commie compute: California Gov. Gavin Newsom proposes turning AI companies' valuations into a public equity fund to retrain displaced workers ahead of a possible 2028 run - Politico

Patch-22: AI could unleash a flood of new vulnerabilities beyond the 100-plus disclosed daily, straining banks and hospitals that can't patch quickly, CrowdStrike warns - Fast Company

Intuit, Technion, and Tel Aviv University researchers identify HalluSquatting, an attack exploiting LLM hallucinations to hijack tools like GitHub Copilot. - TechRadar

The bot with your bank PIN: 1password integration lets Claude sign into websites with your credentials. - 9to5Mac

Risky for commerce and financial sites, you could get prompt-injected and have an agent buy stuff and ship it anywhere, or transfer your cash to a casino in the Philippines, also media websites might block your bots and your whole account.

Predict and serve: Police departments increasingly buy AI tools for facial recognition, gunshot detection, and predictive policing, despite doubts about their effectiveness - The Verge

Digital dirt/cringe check: Employers now use AI to dig up compromising online histories on job candidates, and scrubbing a digital trail can itself backfire - The Wall Street Journal

Bot scrum master: Coders now direct multiple Claude sessions in parallel instead of writing code by hand, reshaping software engineering more than any other profession - Bloomberg

AI coding tools let developers ship faster, but 75% don't fully trust the code and fear losing debugging skills - TechRadar

The great workaround: Shadow AI often appears because approved workplace tools are slow or limited, not because employees are dodging policy - TechRadar

Congratulations, you just hired a million bad employees - A16Z

Fork around and find out: Linus Torvalds tells Linux contributors opposed to AI tools they can fork the kernel or simply walk away. - theregister

Researchers' own analyses of computer-architecture papers lost out to a multi-agent LLM pipeline called Gauntlet in 15 of 20 comparisons, especially on rigor - arXiv.org

GPT-3.5, GPT-4 and GPT-4o skew toward Western moral values, care and consequentialism over purity and deontology, underweighting values in nations like Morocco and Nigeria per over 90,000 real respondents - Fast Company

The Oversight Board tested 10 AI models and found they favor speech-permissive governments while discouraging protest against restrictive ones, a statistically significant pattern - Engadget

Game theory of mind: Game designer says games reveal how minds work, offering a lens on LLM cognition as more games are built for AI to play - Financial Times

AI outsmarted by a toddler: Top AI models struggle to match infant learning on real-world video, per a new Meta/Stanford EgoBabyVLM benchmark - WIRED

WSJ calls BS on PBCs. - The Wall Street Journal

There are many institutions with defined missions, foundations, universities. Mozilla, Hershey, Bosch, Novo Nodisk, Carlsberg, Patagonia are under foundations.

It's like, if the government tries to make things better, that's bad (despite in theory democratically aligned governed).

If a private entity wants to do good things, it's also bad.

Good is bad, up is down, everything's turned around. Black is white, hot is cold, and nothing's quite the way it's told.

Alignment and governance are always hard. As in everything else, there are people who try to keep themselves honest accountable and transparent, and people who shamelessly game the system, and people who criticize and boo from the sidelines.

England had 10 men against Mexico and it was less one-sided.

Tuchel is toast

1) never start a land war in Asia
2) never go up against a Sicilian when death is on the line
3) never park the bus against Argentina, Messi will steal it

undoubtedly 2 great and very different teams in final

Follow the latest AI headlines via druce.ai on Bluesky