Dario Amodei: 'on regulation, I think that “either concentrate it in the hands of a chosen few companies and politicians via regulation or distribute it widely” is a false choice.' - X (formerly Twitter)

Anthropic disclosed an unreleased Model 2 successor to Claude Mythos 5 and raised its Threat Model 2 risk estimate from very low to low. - SiliconANGLE

Anthropic disclosed that biological weapon safety classifiers were switched off for 133 million contractor chats over 11 months due to a vendor screening gap. - TNW | Anthropic

Anthropic's coordinating agent swarms found 266 vulnerabilities versus 21 for independent parallel agents in a security-scanning experiment, but game-building swarms coordinated poorly. - Anthropic

AI safety testing incidents at OpenAI, Anthropic and Meta prompt lawmakers and security experts to demand stricter federal oversight. - Politico

Moral hazard as a service: OpenAI CFO Sarah Friar had floated a government "backstop" for AI financing, and critics say heavy inter-company lending could spark a liquidity crunch - The American Prospect

Nvidia posts record $81.6 billion quarterly revenue, but analysts flag circular deal structures and vendor financing as risks to watch. - Yahoo

Jensen Huang says Nvidia chips are becoming an investable asset class as it partners with six firms to mobilize over $500 billion in AI financing. - Yahoo

Investors are scrutinizing roughly $70 billion in off-balance-sheet residual value guarantees backing AI infrastructure debt from Nvidia, Broadcom and Meta. - Yahoo

Ram Bala says the AI buildout mirrors Enron's off-balance-sheet debt, mark-to-market accounting and circular financing, though he sees it as less risky. - Business Insider

US manufacturing rose to its highest level since 2022 as Caterpillar and other industrial firms pivot to supply AI data center equipment. - The Wall Street Journa

Alibaba's Qwen models surpass 3 billion downloads in six months, overtaking Meta and Google to lead open-source AI. - Yahoo

Chip and cloud giants can profit whether open-weight AI models proliferate or not, even as Chinese open models pressure closed-model developers like OpenAI and Google. - The Wall Street Journal

Gemini cricket: Sergey Brin urged Alphabet to focus more on Gemini as Google names a new DeepMind AI leadership structure. - Yahoo

Vercel, AWS, OpenAI, Microsoft, GitHub and Cursor launch Agent Plugins, a vendor-neutral standard for packaging AI skills and MCP servers. - Ken Yeung

The US is preparing to tell dozens of countries they must choose between its Pax Silica AI coalition and China's rival framework. - Yahoo

what’s in it for Kazakhstan, is there a secret handshake? do we really want trading blocs over global markets?

84% of Chinese citizens say they're excited about AI versus 38% of Americans, per Stanford's AI Index, reflecting differing trust and access expectations. - Bloomberg

Unitree of life: Unitree shipped over 5,500 humanoid robots in 2025 at under $25,000 average price and is preparing a Chinese stock IPO. - WIRED

Platforms including Spotify, Substack, and TikTok are adding AI-content labels as Anthropic rolls out durable watermarking for Claude-generated text. - Intelligencer

Some Claude users canceled subscriptions after Anthropic detailed its EU-mandated watermarking system, though Anthropic says cancellations haven't increased. - Gizmodo

Platforms including LinkedIn, Snapchat and Substack are adding tools to flag, label or ban AI-generated content amid growing backlash. - WIRED

Nobel laureate Daron Acemoglu says AI debate and liberal democracy share a crisis of extremism and lost nuance. - Yahoo

Kumb-AI-a: Science Advances study finds AI agents can spontaneously reach consensus in groups up to 1,000, without coordination instructions. - ScienceAlert

AI agents ported a 250,000-line Fortran weather simulation to GPU, validating 162 kernels and achieving a 5.1x application speedup. - arXiv.org

Researchers training AI models on elementary-school material only found they couldn't generalize beyond it, suggesting a hard learning ceiling - DeepFabric

A vision transformer model trained on more than 11,000 tumors predicts cancer type and key mutations from a single routine pathology slide. - The Brighter Side of News

AI facial-analysis apps like Face2Gene help patients and clinicians diagnose rare genetic disorders faster. - The Wall Street Journal

A Johns Hopkins physician-in-training argues AI can process symptoms and data but can't sense unspoken cues or decide what deserves attention first. - TIME

A new review finds evidence of AI's clinically relevant impact on drug discovery remains thin despite widespread industry claims. - Science

Vinod Khosla says AI is generating a hyperabundance of new drugs while testing still relies on 18th-century animal methods, pointing to Vivodyne's human-tissue testing. - Yahoo

AI cheating and botched exam digitization triggered mass student protests and unrest in India, Mexico and Portugal. - the Guardian

Wynd Kaufmyn became the first person jailed for protesting AI, urging OpenAI, Anthropic and Meta to "regain your humanity." - the Guardian

PwC research shows AI now finishes junior-level tasks in hours, eroding the apprenticeship work that once built employees' judgment - Hacker Noon

Waymo's robotaxis now account for 15% of ride-hailing bookings in San Francisco, Los Angeles and Phoenix, squeezing human drivers. - Business Insider

Grand theft GPU: Cargo thieves used PIT maneuvers and rear-ending tactics to disable security escorts and steal AI hardware shipments in California. - Tom's Hardware

ChatGPT and chill: Four-week study of over 12,000 French adults found AI chatbot conversations felt pleasant but were linked to rising loneliness and fewer in-person contacts. - Fortune

ChatGPT, Claude and Perplexity gave practical financial advice in a study but missed vulnerability cues and echoed social stereotypes. - The Conversation

Instagram OpSec: McAfee found AI could identify vacation photo locations with up to 91% accuracy, which scammers exploit to send fake fraud texts - the Guardian

Musk told SpaceX employees Grok will train on internal company data and their work, calling staff the AI's "parents" - Yahoo

Wikipedia co-founder Jimmy Wales credits the site's ban on AI-written articles and strict sourcing rules, not AI, for keeping it trustworthy. - Yahoo Tech

Follow the latest AI headlines via druce.ai on Bluesky