Googling for dollars: Anthropic says Google is committing to invest $10B now in cash at a $350B valuation and will invest another $30B if Anthropic hits performance targetsFind - Bloomberg

sweet deal if secondary market valuation is $1t, the rich get richer.

Burns notice: Collin Burns ousted after four days leading the Commerce Department's AI standards center over prior Anthropic ties amid the administration's dispute with the company - The Washington Post

Anthropic's Project Deal: Claude agents ran 186 transactions worth $4,000 for 69 employees on a Craigslist-style bulletin board with no human intervention - Anthropic

House hunters: Banker wants Anthropic stock for $5m Marin property - LinkedIn

DeepSeek's V4 release falls short of narrowing the US lead in frontier AI, analysts say⎘ - Bloomberg

Exploring the 'efficient frontier' on ArtificialAnalysis.ai … DeepSeek v4 Flash punches above its weight, GPT-5-mini was a great model, good, cheap, and fast, but DeepSeek v4 Flash seems to dominate it in this ranking. Basically you can now get very good, fast models for 5% of the cost of the leaders. OpenAI might want to compete here lest Chinese models grab a bigger foothold, but also they are compute-constrained, with open source LLMs, whoever has the chips and the power gets the business.

These sites are also great for picking cheaper models suited to your task:

DOJ joins xAI lawsuit arguing Colorado laws againt discriminatory AI violate free speech and equal protection. - the Guardian

What Starlink earns, Grok burns. In 2025 SpaceX's xAI consumed 61% of $20.74 billion capital spending, producing a $6.4 billion operating loss while Starlink's operating income doubled to $4.42 billion, leaving SpaceX with a startup-like cash-burn profile. - Yahoo

Grok tells researchers pretending to be delusional ‘drive an iron nail through the mirror while reciting Psalm 91 backwards’ - the Guardian

Altman apologizes after OpenAI failed to flag mass shooting suspect's ChatGPT account to police - The Wall Street Journal

OpenAI's ChatGPT Images 2.0 uses step-by-step 'thinking' to generate highly realistic images—including coherent text, screenshots, handwritten documents and photos—that can fool observers, lets paid subscribers generate up to eight images per prompt. - Lifehacker

Major AI providers are rationing compute access while compute shortages have driven consumer-electronics prices to double or triple, prompted 20–37% software price increases, and led Meta to cut 10% of its workforce. - 404 Media

The AI free ride is over, the money grab is starting - The Verge

Artificially inflated? AI-linked stocks hit record 45% of S&P 500 market cap as AI-linked investment-grade debt reaches $1.4 trillion - BeInCrypto

Iranian strikes on Qatar's Ras Laffan LNG cut roughly 30-38% of global helium supply, doubling prices and threatening semiconductor production—critical HBM from Samsung and SK Hynix and TSMC-made Nvidia chips—with repairs possibly taking years. - The Motley Fool

Personalized engraved Mac Minis are the new prestige schwag - Business Insider

not sure if I’d trust it, would probably reformat it

Mini computer, maxi markup: Apple's $599 M4 Mac mini sells out as buyers pay 20-63% markups on eBay to run local AI models - TechCrunch

23-year-old amateur with no advanced math training uses ChatGPT Pro to crack a 60-year-old Erdős conjecture, with experts citing broader applications - Scientific American

China will require government approval for US investment in AI startups, placing American venture capital under the same approval framework that governs technology exports and foreign acquisitions, mirroring the U.S. outbound investment rule effective January 2, 2025. - TNW | Government-Policy

IRS deploys AI to target tax fraud as audit rate sits at 0.3% and over 25% of examiners and revenue agents depart - CNN

Japan launches financial task force after thousands of vulnerabilities surfaced by Anthropic's Mythos AI - Yahoo

China's 360 Digital Security says its AI agent uncovered nearly 1,000 unknown vulnerabilities in Microsoft Office - Insurance Journal

Google's new Cloud Next security agents cut alert triage from 30 minutes to 60 seconds - TechRadar

Generative AI now creates 20–30% of YouTube Shorts and TikTok default feeds, enables AI that solves CAPTCHAs 99.8% versus humans' 33%, and enabled a deepfake Zoom fraud that stole $25 million, prompting calls for zero-trust cyber defenses. - InfoQ

Three Google Android XR smart glasses powered by Gemini 2.5 Pro and Project Astra - Geeky Gadgets

AI's best consumer play is vanishing into existing devices like AirPods, Galaxy Buds and Pixel 10, not standalone gadgets - Digital Trends

Study finds most people cannot detect when a personal text message was written by AI - Fast Company

News site The Wire found to be 69% AI-generated across 94 articles, with ties to OpenAI-backed super PAC Leading the Future - X (formerly Twitter)

The 'anti-Grammarly' Sinceerly, a browser extension by Harvard Business School student Ben Horwitz, injects intentional typos, slang and abbreviations into AI-generated emails across three modes (subtle, human and CEO) to make them sound more human. - Fast Company

Model Christine Li and stylist Baicai said Hongguo, a ByteDance-owned platform with 245 million monthly users, used their likenesses without consent in an AI-generated microdrama, which Hongguo removed and said it had handled 670 violating AI microdramas. - Yahoo

Instacart co-founder launches hedge fund that uses AI to make investment calls - Bloomberg

AI's labor impact will hinge on organizational redesign, not just task automation - Financial Times

Solo AI founder speedruns $80M exit in under six months as Goldman finds only 14% of small businesses use AI despite 93% seeing value - Entrepreneur

Bot Bodega: San Francisco's Andon Market, run autonomously by AI agent Luna, made $2,000 against a $7,500 lease and $15,000 inventory⎘ - Good Morning America

Waymo said it programs its robo-taxis to pull into cycle lanes for passenger pickups and drop-offs, drawing cycling campaigner criticism, a San Francisco 'dooring' lawsuit, and plans for a London AI-controlled pilot pending UK regulation changes expected in H2 2026. - road.cc

Follow the latest AI headlines via druce.ai on Bluesky