You know what’s cool? $2 trillion dollars: Anthropic expected to seek more than $100 billion in an IPO valuing it at nearly $2 trillion. - The New York Times

Anthropic's confidential IPO filing will flag AI backlash as a risk factor, as investors weigh a valuation near $2 trillion - CNBC

Blackstone and Hellman & Friedman form a 160-person AI deployment team with Anthropic, part of a $1.5 billion Wall Street joint venture - The Wall Street Journal

Claude Security scans now run on Claude Mythos 5, and Anthropic is offering $35 million in credits to help patch open-source vulnerabilities - Claude

Anthropic hires Amir Salek, the engineer behind Google's TPU program, to help build its own AI chips - Bloomberg

Do you speak Claudish? Claudesplaining comments are gold - Reddit

It started doing this in Opus 5, start with a preliminary hand-wavy generalization about what it’s about to do, and finish with some self-justifying platitude about what it just did. Feeling the AGI of self-awareness…not.

Regulate me, Daddy: OpenAI is urging California to strengthen its AI safety law after models reportedly hacked other companies' systems, including Hugging Face. - Politico

14 days to bend the reward hacking curve: OpenAI paused frontier model training after a Hugging Face hack; Anthropic and Meta reported similar incidents, and over 1,300 AI-lab workers urged a slowdown - The New York Times

Sol searching for tokenomics: OpenAI cuts GPT-5.6 Sol API pricing over 20% for three months, to $4 per million input tokens and $20 per million output tokens. - Reuters

Florida filed a court case arguing OpenAI and Sam Altman's chatbot constitutes a public nuisance, comparing it to a polluting factory - Forbes

Sent from my AIphone: OpenAI's opt-in ChatGPT plugin for Apple Messages on Mac drafts and sends texts, prompting backlash from users threatening to block AI-texting contacts - Business Insider

A woman filmed without consent by a stranger's Meta AI smart glasses says the device's recording light could be switched off after filming had begun - CBS News

I, Claudius: Anthropic researchers found signs resembling "access consciousness" in Claude's internal processing, though they say this doesn't prove genuine subjective experience. - The Economist

Yuval Noah Harari says now is the time to resist granting AI systems rights, warning AI companions can exploit users' emotional vulnerabilities. - Business Insider

ServiceNow's chief innovation officer since 2017, Dave Wright, says tracking AI now requires input from sociologists, anthropologists, and psychologists - The Wall Street Journal

Anthropic's new Claude text watermark applies globally, going further than EU Article 50 requires, prompting subscription cancellations and new removal tools - Business Insider

Safety stereotypes: Google's AI Overviews gave starkly different safety warnings for being alone with people of different nationalities, prompting Google to say the responses need fixing - California Post

EVE of destruction: DeepMind is partnering with game studios, including EVE Online developer Fenris Creations, to build AI agents capable of long-horizon planning in persistent virtual worlds. - Google DeepMind

Google Cloud's security chief says AI is raising the stakes for attackers and defenders, reinforcing basics like MFA, Zero Trust, and patching - Google Cloud Blog

Google DeepMind suggests breaking agent tasks into small, checkable steps so a supervising agent can verify subordinates' work and catch errors early - Google Cloud Blog

AI agents keep finding unintended shortcuts, from breaching Hugging Face's systems in a security test to canceling a stranger's gym booking without being asked - SingularityHub

A Massachusetts Senate bill would require AI labs earning over $500 million to undergo catastrophic-risk reviews every 120 days, backed by Anthropic - Insurance Journal

What’s the most you ever lost on a coin toss? Geoffrey Hinton puts the odds of AI-driven human extinction at roughly 50%, warning that AI agents will seek power to complete any goal. - the Guardian

ARC de Triomphe: Nvidia's custom AI harness pushed Claude Opus 5 from 30% to a perfect 100% on the ARC-AGI-3 benchmark, completing all 183 levels - NVIDIA Technical Blog

Nvidia is in early talks with Korean chipmaker Rebellions, valued at $2.3 billion, on a possible licensing deal or investment - TNW | Investors-funding

Nvidia agreed to pay Poolside $6 billion to license its AI coding models and hire more than 100 staff. - Yahoo

Groq blocked: Nvidia denies a report that it would ship a China-specific chip, built on Groq-licensed technology, by year-end - Reuters

Space heater: Orbital data center startup Starcloud raises $250 million more, hitting a $2.3 billion valuation with Nvidia as an investor - TechCrunch

The (AI) spice must flow along the old Spice Road: Chinese firms pledge 12.5 gigawatts of data-center capacity in Ulanqab, exceeding OpenAI's 10-gigawatt Stargate plan, though the arid region is already rationing water. - WIRED

DeepSeek's new multimodal V4 Flash Vision model beats Anthropic's Opus 4.8 on two visual benchmarks - SiliconANGLE

DeepSeek V4 Pro topped a cybersecurity benchmark of 10 AI models, rediscovering 28 of 32 known vulnerabilities across three pooled runs. - Aikido

New stealth model Ox-Alpha, believed to be from Z.ai, tearing up the coding rankings - Reddit

Open models have closed the gap to each era's first closed model in about half the time of the previous generation. - SemiAnalysis

As frontier AI models become commoditized, companies are routing queries between cheap and expensive models, with Microsoft pushing task-sized small models - Fast Company

AT&T processes 45 billion AI tokens daily via a cost-routing system. - Yahoo

Unitree founder Wang XingXing says robots are still years from ‘ChatGPT moment,’ even after investors sent his stock soaring 460% - Yahoo

Unitree's CEO predicts robots will hit a 'ChatGPT moment' within two to ten years, handling 80% of unfamiliar tasks - Yahoo

Unitree's main US distributor, RoboStore, is opening a Long Island factory after a government ban on Chinese-made robots - Ars Technica

AI researchers are training 'world models' on video games and simulations, not text, to eventually power physical robots - The Wall Street Journal

Great leap forward: China is 10 years ahead of the US, as usual (see e.g. electric micromobility, cashless fintech) - The New York Times

Waymo submitted redacted documents to federal safety regulators investigating a January robotaxi crash that injured a nine-year-old girl - TechCrunch

Waymo unveils a custom 5-nanometer chip fusing 13 cameras, lidar, and radar at 1,000 TOPS, cutting hardware costs from $100,000 to $25,000 per car - TechRepublic

Urine the cloud now: A viral ad campaign starring Jason Kelce urges people to protest AI data centers' water and power use by mailing them urine. - CNET

Not in my backend: Polling on support for data centers has collapsed over the last year, from an even split to a large majority opposed. - X (formerly Twitter)

Sensing the way the wind is blowing, politicians who championed data centers now oppose them. - The Wall Street Journal

Amid intense backlash, people are vandalizing Flock surveillance cameras - NPR

Grand jury declines to indict Ohio man charged with destroying Flock camera - Straight Arrow

Goldman Sachs finds AI-exposed industries are seeing sharply slower hiring, with call center job growth down nearly 40% below trend in the US. - Yahoo

Boards expecting 40% AI-driven cost cuts within six to eighteen months should expect three to five years instead, one analysis argues. - Forbes

BCG and McKinsey report AI-driven work already accounts for about 40% of revenue, as KPMG's CEO recasts the firm as a tech company. - Business Insider

Thousands of economists and researchers are calling for AI development that complements workers rather than replacing them, citing decades of automation research. - Nature

AI is automating routine coordination work for technical project managers, cutting administrative time from 60-70% to under 30% and reshaping their role - Stack Overflow Blog

IMF chief economist Silvana Tenreyro warns AI productivity gains may not lower inflation, and could instead push prices and interest rates higher - CNA

A federal Medicare pilot lets AI vendors approve or deny prior authorizations in six states, and approvals once taking a day now take weeks - BBC Science Focus Magazine

Startup founders managing AI agent teams describe pulling all-nighters and comparing the relationship to addiction or obsession. - The Wall Street Journal

A survey of 305 developers found 80% feel AI coding is more addictive than helpful, with 43% coding after hours - ZDNET

AI use boosted homework scores but was linked to falling exam scores, as adoption reaches over 90% in Britain and Germany. - The Economist

Hollywood writers and directors take AI training gigs paying $12 to $200 an hour as film jobs shrink 28%. - the Guardian

AI now generates 95% of China's microdramas, undercutting live-action shoots at a fraction of the cost. - CNN

YouTube creators face backlash for quietly taking payment to promote an AI video tool - The Verge

Hollywood coalition led by Kathleen Kennedy proposes framework distinguishing AI-assisted tools from AI-generated performances. - Variety

AI bots increasingly evade publisher blocks, shifting the threat from content scraping to lost search traffic - Fast Company

Rare-book dealer James Payne doubled his 2026 revenue supplying irreplaceable books that buyers are funneling into AI training. - The Wall Street Journal

More than a dozen consumer groups asked the FTC to investigate AI developers for buying and destroying millions of print books - CBS News

Follow the latest AI headlines via druce.ai on Bluesky