Geneva convention for GPUs? Demis Hassabis proposes a US-led coalition to review advanced AI models before public release, calling AGI a few years away. - Substack

No Apple-y ever after for OpenAI: Apple's lawsuit accusing OpenAI of stealing trade secrets threatens to disrupt OpenAI's device ambitions before the case is resolved. - Bloomberg

OpenAI's hiring of over 400 Apple employees, including Jony Ive, breaks Silicon Valley's usual talent-poaching norms, angering Apple. - Business Insider

Search results with a side of gambling addiction: OpenAI now integrates Kalshi's prediction-market odds for World Cup matches directly in ChatGPT search results. - The New York Times

Prompting guide for ChatGPT 5.6 - OpenAI

Peak AI: ChatGPT's suggested shortcut stranded two hikers on a Polish mountain traverse, requiring a helicopter rescue. - Digital Trends

Failing Sesame Street benchmarks: Influencer Husk says OpenAI's new GPT-Live voice model miscounts the letter E in "seventeen." - Fast Company

Token shrinkflation: Analysis finds Claude's newest tokenizer produces up to 73% more tokens than GPT's for identical code, raising effective costs despite similar list prices. - Playcode.io

Anthropic analyzed the personality and values of the different Claude models. - Anthropic

A hands-on review calls Claude the most comprehensive AI toolset, citing broad MCP support, million-token context windows, and a persistent memory feature. - BGR

Satya Nadella says AI buyers pay twice, in cash and in proprietary data leaked through usage, in a viral essay. - TNW | Artificial-Intelligence

In theory orgs have zero-day retention agreements and no training clauses in place tho? Just knowing the use cases that enterprises pay a lot of money for is gold.

OpenAI asked a judge to rule xAI's trade secret lawsuit meritless and seeks over $1 million in legal costs. - Bloomberg

Commit crimes no more? AI appears to no longer upload entire Git repos to a GCP bucket. - International Cyber Digest

xAI installed 59 unpermitted gas turbines for its Colossus 2 data center, disproportionately polluting Black communities in Mississippi. - Yahoo

OpenRouter data suggests Grok 4.5 could gain AI spending market share in coming months. - NextBigFuture.com

Office SpaceX: Cursor is developing Sand, an office AI agent rivaling Claude Cowork, as SpaceX preps a $60 billion buyout of parent company Anysphere. - TNW | Anthropic

Elon Musk loses the plot: Musk says SpaceX could be worth more than Earth as the company unveils its first AI compute satellite, AI1. - Yahoo

Beyond-the-clouds computing: Most experts are skeptical of data centers in space - Yahoo

The Great AI Power Grab: Georgia Power is acquiring over 300 land parcels, including homes, via eminent domain for a transmission line serving new AI data centers. - CBS News

Meta's Hyperion data center budget jumped from $10 billion to over $50 billion in two years, straining its Louisiana host town with rents and evictions. - Yahoo

Power transformer delivery times have stretched to 24-48 months and prices have risen 50-80%, driven by AI data centers, EVs, and aging grid infrastructure. - TechRadar

New York became the first US state to halt new data center construction for a year, citing rising energy prices and water use concerns. - The Washington Post

SoftBank's Masayoshi Son says nuclear fusion will eventually power AI data centers as demand reaches 3 terawatts by 2040. - Bloomberg

Mistral CEO Arthur Mensch warned European officials and AI leaders that France should not sell its cheap nuclear power advantage to American firms. - TNW | Artificial-Intelligence

AI data center construction workers are filling Iowa campgrounds, displacing recreational campers as public opposition to new AI facilities grows nationwide. - BGR

Silicon curtain: Nvidia has halved the list of approved Asia buyers for its chips as Washington tightens its crackdown on exports to China. - Financial Times

A thriving black market is helping advanced AI semiconductors slip past US trade barriers and reach buyers in China. - Financial Times

Chinese startup Dongfang Suanxin unveils DF1000 AI chip built on a fully domestic supply chain to bypass US tech restrictions. - The Wall Street Journal

Saudi Arabia's Humain and the UAE's G42 are building massive Nvidia-powered data centers, but US export rules still limit their chip independence. - Rest of World

The Trump administration and industry groups are weighing a framework tying US open-source AI model releases to China's capability level. - The Washington Post

Her, eventually: iPhone users can join Apple's Siri waitlist in Settings to access the new Siri AI in iOS 27's public beta. - BGR

Apple's first public betas of iOS 27 and macOS 27 debut a conversational Siri AI, launching English-only and unavailable in the EU. - Engadget

Apple's revamped Siri AI, unveiled at WWDC, is now available to iPhone owners via the iOS 27 public beta. - WIRED

Siri AI is changing how we use the iPhone - The Verge

Montefiore Medical Center laid off 12 utilization review nurses for AI software, weeks after a strike won contract protections against automation. - TechSpot

Mark Cuban warns AI-armed insurers could delay and deny care, while providers and insurers both use AI to manipulate contracts and revenue cycles. - Business Insider

Foreign fraud rings use AI voice bots to impersonate healthcare workers, bilking Medicare of billions via a $10 billion catheter scheme. - New York Post

Negligence as a service? Wrongful-death lawsuits against OpenAI and Google, plus a Florida criminal probe into ChatGPT, test whether liability law can handle autonomous AI agents. - Yahoo

The robot did it: As businesses shift from chatbot errors to autonomous AI agents, they face a widening range of potential legal claims. - Financial Times

A federal judge ruled OpenAI users can't quash search warrants for chatbot histories, exposing a new AI privacy risk under the Stored Communications Act. - Above the Law

HUD denied a FOIA request for over 100 documents on DOGE's AI use in housing policy, citing a nonexistent 'AI privilege.' - WIRED

Make love, not AGI: About 200 protesters marched between OpenAI, Anthropic, and Google DeepMind offices in San Francisco demanding a pause on frontier AI training. - Decrypt

ACRouter, an AI agent that routes each coding task to the best model, beat all-Opus setups by 2.6x on cost in a 10,000-task benchmark. - VentureBeat

Building AI Agents? Here Are Some Anti-Patterns to Avoid. - MachineLearningMastery.com

Mariner, a $630 billion RIA, signed a five-year deal with Humanity Labs to deploy an AI workforce of 700+ FTEs. - Wealth Management

Enterprise AI is shifting from simple chatbots to multi-step workflows that connect models with business data, tools, and permissions. - Hacker Noon

UK businesses layering AI onto fragmented systems face a 'memory crisis,' as most AI tools reconstruct context from scratch each time. - TechRadar

BMO's chief AI and data officer describes agent governance and performance evaluations the bank uses to build trust in AI models. - American Banker

Interview with the AI vampire? New York nonprofit's new hire aced the interview but froze on basic tasks, leading the manager to suspect the hire used a chatbot. - Bloomberg

Workers say bosses credit AI for their actual work, and also penalize them for disclosing AI use, researchers call it the AI penalty. - TNW | Amazon

The Liang King: DeepSeek founder Liang Wenfeng's net worth doubled to $36 billion, making him the richest AI model creator - Bloomberg

San Francisco police drone footage leaked - WIRED

LAPD wrongly investigated 161 people over falsely flagged stolen-vehicle alerts in a two-month audit of its license plate readers. - 404 Media

Let us use it, or lose it, Samsung will delete health data like sleep, medication, and cycle records unless users agree to let it train AI models. - Neowin

Context bomb squad: Security teams now plant decoy "context bomb" prompt injections that trip AI hacking agents' refusal mechanisms, cutting successful account takeovers from 57% to 5%. - Ars Technica

Experts from MIT, Runway, and World Labs discuss the promise and limits of world models as an alternative to large language models. - Ars Technica

AI discourse is shifting from prompts to self-improving loops, which the author says require corporate governance via a self-model. - Fast Company

Comma AI's George Hotz argued locally controlled, user-aligned AI should not refuse requests, comparing it to a gun that doesn't judge use. - TechCrunch

What if it’s CSAM, bioweapons, making fake docs to defraud tho.

Follow the latest AI headlines via druce.ai on Bluesky