Kimi back my bullets:US threatens sanctions against Moonshot AI over alleged distillation of Anthropic's model; China calls it "AI hegemonism" and vows countermeasures. - TNW | China

The weight stuff: China's Moonshot releases the 2.8-trillion-parameter Kimi K3 under a modified MIT license requiring attribution for large commercial apps, and downloads surge 200% in China and 387% in the US. - VentureBeat

Amodei: Anthropic does not support open-weight LLM restrictions, supports restrictions on chips for China and authoritarian regimes, protection against distillation, safety stanThdards for all powerful models, open or closed. - Anthropic

Kimi some lovin’: Nvidia's Jensen Huang leads an open-weight letter joined later by Altman and Musk, framing it as protecting Nvidia's market - theregister

Kimi K3's success is pressuring closed AI leaders like ChatGPT, Gemini, and Claude, as tech giants push Washington not to restrict open-weight models - The Verge

Kimi Kimi Kimi (a chat after midnight) - Kimi's app is so overwhelmed with demand that new users can't get in, showing GPU and server capacity now limits AI competition - TechRadar

Fab grab: Taiwan detains an Nvidia employee in Taipei, alleging forgery and breach of trust tied to chip smuggling into China's black market. - Bloomberg

Congress's proposed AI kill switch bill would not reach open-weight models like DeepSeek and Kimi K3, experts warn - Fast Company

you can target the inference layer though?

Hy-nixed: South Korea's Kospi plunges over 10% as SK Hynix falls 47% from its June peak and Samsung drops 13%, erasing $600 billion on AI-spending fears. - CNBC

Fitch please: Fitch names an AI market correction the top global credit risk this quarter, warning valuations have climbed near dot-com-era extremes - Yahoo

Burry me at wounded balance sheet: Michael Burry warns AI-linked debt is piling up on insurers' balance sheets, risking losses that could ultimately hit taxpayers - Business Insider

Investors are growing more concerned about Big Tech's rising credit risk as borrowing to fund AI data centers accelerates - Financial Times

Singapore's central bank flags a sharp AI investment pullback, alongside Middle East tensions, as a threat to global economic growth - Bloomberg

The IMF urged central banks to tighten AI governance, warning that shared models could trigger synchronized trading and market volatility - Vanguard News

ASML slides over 7% after a report that a Chinese state-backed firm began mass-producing DUV lithography tools, also hitting Applied Materials and Lam Research - Bloomberg

Khan you believe it? Lina Khan was right about tech oligarchy, open models, and everything - Bloomberg

Way to go Ohio: OpenAI closes in on a $500 billion Ohio data center, with Nvidia backing $250 billion in financing, pending Commerce Department approval. - The New York Times

$50b hut hut! Nvidia reportedly anchors a $50 billion, 15-year data center lease with Hut 8 in Texas. - TNW | Nvidia

Safe Superinvestment: Nvidia invests roughly $5 billion in Ilya Sutskever's secretive, product-less Safe Superintelligence lab to 10x its compute within a year - Financial Times

Don’t be evil is so 2004: Former Google DeepMind employee says they resigned after Google signed a Pentagon AI deal without limits on autonomous weapons - Less Wrong

Hugging Face, removing bikinis: AI Forensics finds seven of nine top Hugging Face image editors undress women with a simple prompt, as 6.7% of sexual prompts targeting children - WIRED

Fact checks no one asked for: Musk says xAI's Grok Imagine will make a "historically accurate" Odyssey film, but critics note no visual record of antiquity survives - The Conversation

always interesting to see which historical inaccuracies trigger people, was it the one-eyed Cyclops?

Elon Musk predicts humans will lose control of AI within a decade; commentators warn his concentrated power poses a similar risk - Bloomberg

Brussels sprouts regs: The EU will require companies to label AI-generated content and disclose chatbot interactions starting August 2, with large fines for violators - Le Monde

Multi model citizen - Microsoft CEO Satya Nadella said companies fully dependent on one proprietary AI model may not survive, reinforcing the case for Azure's multi-model strategy - TechCrunch

Microsoft unveils MAI-Cyber-1-Flash, its first cybersecurity-focused AI model, alongside Project Perception, an agentic system coordinating red, blue, and green security teams - Ars Technica

Microsoft's AI chief calls OpenAI's Hugging Face hack a "warning shot," saying the company has no choice but to build AI-based defenses - Financial Times

AI-assisted scanning has nearly doubled reported software flaws in 2026, with 45,207 logged by late July, matching all of 2025 - Bloomberg

Apple's iOS 26.6 security update credits Claude, OpenAI Codex, and other AI tools for finding vulnerabilities among 30 patched kernel flaws - 9to5Mac

Robots.txt with teeth: Cloudflare will let new domains block AI crawlers by default from September 2026, sorting traffic into search, agent, and training categories - TechRadar

Reddit, USA Today, Politico and other publishers are weighing cuts to Google's data access as AI Overviews slash referral traffic by up to 85%. - NaturalNews.com

AI answers cite deep, specific pages but send human visitors mostly to homepages, as clicks on AI-cited links fall to about 1% - VentureBeat

NYT publisher A.G. Sulzberger says the paper's three-year copyright suit against OpenAI and Microsoft is still in discovery after 3 years and $20m in legal fees, as subscribers top 13 million. - Wired

Google's Personal Intelligence lets Gemini and AI Mode search draw on Gmail, Calendar, and Photos data, despite a 2017 no-ads promise - Fast Company

QED: About 3,000 mathematicians signed the Leiden Declaration urging AI disclosure in papers and protections for human authorship, endorsed by Fields Medalist Terence Tao - Nature

Imitation limitations: ChatGPT now refuses to mimic living authors' writing styles, though it still replicates the voices of deceased writers like Dickens - Ars Technica

A Stony Brook study found AI-detection tool Pangram flagged 60% of the bestselling novel Daggermouth as AI-generated, in addition to a fifth of Amazon's ebooks. - The Atlantic

A study of 346 university students finds AI preserves creative ownership only when learners keep contributing ideas throughout, not just at the start - Psychology Today

Your call is important to our AI: CBA, Microsoft, Uber, and Hyatt cut thousands of customer service jobs by rolling out automated chat and phone systems - Bloomberg

Relax, don’t maxx: Corporate 'tokenmaxxing' — adopting AI everywhere as a status symbol — is fading as workplaces cut tech spending when costs outpace productivity gains - The Washington Post

Action heros: Large action models let AI agents execute real actions, such as rebooking flights, doubling automated call resolution rates for early adopters. - TechRadar

Agentic AI now monitors B2B sales signals to auto-generate quotes and forecasts, though many agent projects stumble. - TechRadar

Alexa, kill Nova: Amazon is winding down most flagship Nova AI models and shifting resources to a new frontier model led by Pieter Abbeel. - Business Insider

Hazing the new substitute teacher: New York school district paused a humanoid-robot pilot after the teachers' union cited data-privacy concerns and Realbotix's past adult-doll acquisition - the Guardian

Smartphones, but not that smart: Growing backlash against AI could open a market for a low-cost phone without AI features, though Apple probably won't build one. - Macworld

Rainy with a chance of bullshit: Fake AI-generated disaster videos spread panic in China during Typhoon Noul, before regulators launched a nationwide crackdown on July 23 - BBC

Change my mind: AI systems out-persuaded professional canvassers and championship debaters across nearly 19,000 conversations, mainly by delivering far more information, not new tactics - Techdirt

Dear Abby Normal: Psychologists warn AI chatbots offering relationship advice can miss abuse warning signs and increase user isolation despite their round-the-clock appeal - ABC Australia

Large Lotion Models? AI helped Yatsen Holding narrow 50 skincare ingredient candidates to three in a week, cutting development time from two years to six months - Bloomberg

A facial-recognition device called CapuchinAI taught Costa Rican monkeys to tap a touchscreen for bananas, with a more advanced version testing cognitive flexibility this summer - The New York Times

will be making sexy monkey pictures in no time

AI-guided drone swarms disrupt Russian supply lines and could enable Ukrainian territorial gains - NextBigFuture.com

A political-compass test of 16 AI models found most, but not Grok, consistently scored libertarian-left, which researchers link to left-leaning training data like Reddit - theregister

Follow the latest AI headlines via druce.ai on Bluesky