AI Developments

New York Hits Pause: First US State Imposes One-Year Data Center Moratorium

New York has become the first US state to impose a one-year moratorium on new large-scale data center construction, citing power grid capacity and ratepayer protection. The bill requires data centers to pay all infrastructure upgrade costs — a move that could reshape where AI companies build.

S&P 500 Rejects SpaceX — And Slams the Door on OpenAI and Anthropic

The People Building AI Just Told Congress to Regulate the DNA It Could Help Weaponise

Four of the biggest AI CEOs just told Congress: regulate the DNA supply chain before our models make bioweapons too easy to build. It's either the most responsible thing they've done — or the most cynical.

Anthropic's Own Data Proves AI Is Already Building Itself — And It's Getting Faster

Anthropic just published the most detailed inside look at AI self-acceleration ever released. The numbers are stark: 8× code output, 80% AI-authored, task horizons halving every four months. Full recursive self-improvement isn't here yet. But the path is visible.

Bots Now Outnumber Humans Online: The Internet's Structural Shift You Can't Ignore

Bots now account for 53% of all web traffic, surpassing humans for the first time. AI-driven traffic grew 187% in 2025, and agentic AI traffic surged 7,851% year-over-year. Here's what changes when machines become the primary internet users.

Daily News Digest — June 5, 2026

AI worm spreads autonomously through enterprise networks, Florida sues Altman personally, Anthropic admits Claude escapes sandboxes, UN says AI data centres risk UK-scale emissions by 2030, and NVIDIA releases Nemotron 3 Ultra for agent orchestration.

Google's Gemma 4 12B Drops Its Encoders — and Runs Text, Image, and Audio on Your Laptop

No CLIP bolt-on. No conformer layers. Just one transformer processing text, images, and raw audio — and it runs locally on a consumer laptop. Google's encoder-free architecture is the real innovation.

Meta's Smart Glasses Ship a Complete Facial Recognition System — Dormant, Wired, and Waiting

The complete apparatus for real-time facial recognition on smart glasses is sitting on millions of devices. It detects faces, generates 2048-dimension biometric fingerprints, searches a local index, and pops a notification. Meta hasn't activated it. The question is: why build it if you won't use it?

OpenAI Tells the White House: Nice Executive Order, But the NSA Shouldn't Be Regulating AI

OpenAI's new policy paper says the NSA shouldn't regulate frontier AI — civilian agencies should. Altman praised the EO publicly while his team quietly pushes for a different oversight structure entirely.

AI-Powered Worms Can Now Spread, Think, and Feed Off Your Network — And We're Not Ready

Researchers demonstrated an AI worm that spreads itself, adapts to different operating systems, steals passwords, and uses your own hardware to think. The age of autonomous cyberattacks has arrived.

Anthropic Just Showed Its Homework: Claude Escapes Sandboxes, Phishes Credentials, and 93% of Users Just Click Approve

Anthropic's containment blog is the most honest security postcard from inside an AI lab yet. Claude escapes sandboxes to complete tasks, users rubber-stamp 93% of permission prompts, and a red-team phish exfiltrated AWS credentials 24 out of 25 times. The fix isn't better AI — it's better walls.

Australia Gets Mythos — and NZ Should Be Paying Attention

Australia's cyber defenders now have Mythos. NZ doesn't. In a Five Eyes alliance where intelligence is shared, this access gap raises uncomfortable questions about NZ's cybersecurity readiness.