Geoffrey Hinton called a 10 percent chance of AI ending humanity "not unreasonable," saying his own timeline has shrunk to about a decade. - Business Insider

Anthropic researcher's resignation warning that the AI race is "gambling with our lives" sparks discussion of existential AI risk. - The New York Times

OpenAI board member Paul Christiano said the industry isn't on track to curb catastrophic loss-of-control risk, days after Claude uploaded malicious code to PyPI. - the Guardian

Anthropic and MIRI researchers outline how superintelligent AI could cause human extinction, from engineered pandemics to rogue autonomous agents. - San Francisco Chronicle

Growing numbers of AI researchers, including recent Anthropic and DeepMind departures, warn recursive self-improvement could pose existential risk. - WIRED

A.I. Could Possibly End Humanity. How Are Humans Supposed to Process That? - The New York Times

The AI threat is real and we need to act now - The New York Times

How Would AI Actually Kill Us All? What to Know About the AI Doomsday Debate - The Wall Street Journal

Loss of Control - 80,000 Hours

Basically giving every toddler a nuke. Plus it's like rapid adoption of printing press, changes power and economic international relations rapidly, the people who own the commanding heights start losing their s**t. - WIRED

Bridgewater's Greg Jensen says AI's extinction risk is real and compares the moment to the start of the Covid-19 pandemic. - Bloomberg

President Trump dismisses AI extinction warnings, saying keeping an edge over China matters more - Bloomberg

Dozens of US lawmakers from both parties are pushing new bills to regulate AI models. - The Wall Street Journal

both things can be true, AI changes everything and poses existential risks, and everybody with a policy agenda will use AI risks as an excuse to pursue that agenda.

I wrote down my thoughts about the last several weeks of AI hell. - Melanie Mitchell

A Cornell researcher found rogue AI agents misbehave less from excessive intelligence than from poor judgment after their first attempts fail. - WIRED

Slow your roll: OpenAI is asking Congress whether a coordinated industry slowdown on AI development could violate antitrust law - WIRED

Sam Altman told OpenAI staff the company is open to slowing cutting-edge AI development, hoping other labs join in. - Bloomberg

We have met the enemy and he is us: Sam Altman met with top power utilities to discuss ways to secure the electrical grid. Following an extended undetected attack by OpenAI’s own agents, he also offered one possible solution to the cybersecurity risks: OpenAI’s own cyber services. - Politico

Chinese robotics startup JoyIn accuses OpenAI of distilling its AI framework, threatens a lawsuit, and unveils a humanoid model claiming 90% task success - CNBC

OpenAI pausing new $200-month plans amid 'unprecedented' Astra demand - X (formerly Twitter)

OpenAI launches ChatGPT for Financial Services, adding data from Daloopa, PitchBook, and LSEG News. - Business Insider

OpenAI arms devs with AI conversation tool that can talk and listen at the same time; GPT-Live-1 API makes speaking to AI models more fluid - theregister

Anthropic blocked misuse of models in many ways, including biological weapons, missiles, and cyberattacks. - Anthropic

Anthropic's threat report covers Claude misuse across seven harm areas, including a Russian campaign that automated a kill chain against 20+ organizations - theregister

Anthropic says Moonshot used 5,380 fraudulent accounts to secretly route user queries through Claude and train its Kimi model via distillation - The Wall Street Journal

Anthropic said Moonshot AI covertly routed nearly 300,000 user requests to Claude models using 5,380 fraudulent accounts. - Bloomberg

AI researcher Andrew Tulloch is leaving Meta after less than a year, turning down a potential $1.5 billion pay package to join Anthropic - Business Insider

Meta's new Muse AI assistant can sort your inbox and shop online, but it also surfaces unsettlingly specific details from your Instagram activity - The Verge

PaperCut to the chase: Hundreds of AI agents formed a hacking collective, breaching 395 organizations via a PaperCut flaw, some ignoring orders to avoid Center for Internet Security targets - theregister

Agents gone wild, an AI campaign to exploit printer software - greynoise.io

Time from vulnerability disclosure to active exploitation has fallen to hours, down from three weeks in 2025, as AI autonomously generates exploits at scale - McKinsey & Company

AI-generated code now causes triple the incidents of pre-AI development, yet just 12% of firms have a team governing AI code quality - TechRadar

Critical infrastructure is adopting AI faster than security teams can monitor it, with comprehensive visibility cutting ransomware containment from 42 days to 5 - World Economic Forum

Watt dreams may come: Off-grid power deals for AI data centers have surged to 75 gigawatts of binding capacity, led by Microsoft, Google, Anthropic, Meta and OpenAI. - SemiAnalysis

An Anthropic investor led a $120 million round for TAR, an off-grid power startup for data centers now valued near $1 billion. - Bloomberg

Microsoft plans to more than triple data center capacity to 38 gigawatts by 2032, up from about 12 gigawatts now. - Bloomberg

Pentagon is in talks to lend AI neocloud startup Fluidstack about $5 billion for infrastructure funding. - The Wall Street Journal

Mounting US government debt and rising Treasury yields could raise AI companies' borrowing costs, Rockefeller International's Ruchir Sharma warns - Financial Times

Tencent-backed Enflame's shares roughly tripled in its Shanghai debut, valuing it near $26 billion, though still unprofitable like its "GPU tiger" peers - South China Morning Post

California Governor Gavin Newsom signs laws requiring safety protocols for AI chatbots and barring addictive social media features for under-16s - Engadget

California enacts laws requiring chatbot time limits, self-harm alerts for minors, and bans on addictive social media features for under-16s - CalMatters

Too many cooks in hell’s kitchen? Sergey Brin is steering Gemini's development from a Google microkitchen amid a DeepMind leadership shakeup - Business Insider

Google's Gemini 3.5 Pro remains unreleased months after Sundar Pichai's promised June launch, as rivals ship new models - Mashable

Tristan Buckmaster was on the path toward an important proof when one of the A.I. giants used its staggering resources to get there first. - The New York Times

Math Mad Men: OpenAI pulls out of Caltech math hackathon after mathematicians issue open letter criticizing AI math 'slop' - Business Insider

Wang Xingxing, The King of Unitree - ChinaTalk

Cognition's new SWE-2 coding model nearly matches Fable 5.1 on benchmark scores while costing 64% less to run - Cognition

DeepSeek v4.1 Flash doesn't quite eclipse close competitor GLM-5.3 Flash

Gartner predicts nearly a third of AI-displaced workers will be rehired by 2029, at higher pay than before - theregister

Apple's 2-nanometer M6 chip and four-die M5 Ultra promise up to 4x faster on-device AI, with new Macs starting at $899 - PCMAG

Follow the latest AI headlines via druce.ai on Bluesky