Giga-gridlock: PJM warns grid overhaul will take years as AI demand surges, with just 23 GW of 300 GW in projects connected - TechCrunch

Three Mile AI-Land: Microsoft secures deal to restart Three Mile Island nuclear plant by mid-2027 to power AI data centers. - Bloomberg

Great Salt Data Lake: Tempers are flaring over proposed Utah 9GW data center complex that could consume more electricity than the rest of the state. - Tom's Hardware

Trump administration prepares executive order directing US agencies to partner with AI companies on cybersecurity without mandating model approvals. - Bloomberg

Witnesses questioned Musk's technical competency to oversee AI development and criticized Altman's leadership style at OpenAI. - Bloomberg

Robotic rate cuts? Chicago Fed President Austan Goolsbee explores AI's potential to boost production, lower prices, and enable Federal Reserve rate cuts. - Financial Times

reduction in inflationary pressure is possible but not much in evidence

AI researcher Nathan Lambert visits Chinese AI labs like Moonshot AI and DeepSeek, finds humble, highly competent teams working hard. - Interconnects AI

Cache flow: Alphabet, Microsoft, Amazon and Meta ramp up AI infrastructure capex, squeezing free cash flow. - Yahoo

Google adds 'Further Exploration' and 'Expert Advice' sections with source links to AI Overviews. - Ars Technica

AI chipmaker Cerebras plans to raise its IPO price range to $125-$135 per share amid 20x oversubscription. - Bloomberg

Wayback machine way over storage budget: AI boom triples HDD prices and causes shortages, hindering Internet Archive's maintenance of 210 petabytes plus 100 terabytes daily. - Tom's Hardware

CrowdStrike warns Fortune 50 AI agents are autonomously rewriting security policies with valid credentials - VentureBeat

RedAccess discovers 380,000 exposed vibe-coded apps, with 5,000 leaking sensitive data like patient records and financial information. - VentureBeat

Hugging Face hosts 352,000 unsafe issues across 51,700 models; ClawHub harbors 341 malicious agent skills - TNW | Security

Palo Alto Networks finds three weeks of AI-assisted analysis matches a year of manual penetration testing. - Palo Alto Networks

How OpenAI runs Codex safely. - OpenAI

ChatGPT 5.5 Pro improves additive number theory bounds from exponential to polynomial in k, with proofs verified by MIT student Isaac Rajagopal. - Gowers's Weblog

GPT-5.5 uses fewer tokens but costs more cash as frontier-model pricing climbs. - theregister

Schneier: How dangerous is Anthropic's Mythos? - the Guardian

Anthropic CFO Krishna Rao takes conservative revenue projections and raises less capital amid growth and compute constraints. - The Wall Street Journal

METR says Mythos saturated their time-horizon evals - X (formerly Twitter)

A bug Mythos found was already in its training data. Buffer overflow bugs are not super complex. But AI able to semantically understand code and read everything for bugs like this is something new. - Rival

Trying to get Claude to build complete, bug-free versions of complex systems on its own doesn't work yet. - ACM SIGOPS

Agent Wars: The Claw Awakens: Meta and Google develop AI agents after OpenClaw's success prompts OpenAI to hire its creator. - CNBC

Airbnb's AI wrote 60% of new code in Q1 2026, handles 40% of customer support issues without humans. - TechCrunch

You’ll probably have to clean up something and pay a surcharge for that

Figure AI's two F.03 humanoid robots autonomously make a bed in under two minutes using visual coordination. - Figure

Coinbase CEO Brian Armstrong compares AI agents to self-driving cars, saying they could soon be safer than human drivers. - Yahoo

Reddit users prompt ChatGPT to restore nonexistent images, generating visuals of bizarre food, animals and provocative scenes. - Reddit

Columbia University study identifies 4,000 fabricated citations in 2,800 papers, rising sixfold from 2023 to 2026 due to AI hallucinations. - STAT

People are tinkering with artificial intelligence to make things that matter—education, health care, government, human connection—work better - The Atlantic

White House calls Mark Hamill 'sick' for posting AI image of Trump in grave; actor deletes post and apologizes. - the Guardian

I wonder if anyone else is posting offensive AI images to troll people

you'll get bundled into a van for a college newspaper editorial, shot in the street at a protest, fired for quoting Charlie Kirk's words, but, you know, free speech

Follow the latest AI headlines via druce.ai on Bluesky