Live Feed
Unitree GD01: World's First Production Piloted Mech Launches at $574,000, Transforms Between Bipedal and Quadruped
Unitree launched the GD01 on May 12, a manned transformable mech priced from 3.9 million yuan ($574,000) that switches between bipedal and quadruped form in seconds. It weighs 500 kg with a rider and is officially classed as a civilian vehicle. The company is also filing for a $620M IPO on Shanghai's STAR Market.
Nscale Closes $790M for Norway AI Data Center as Capital Stack Tops $4B
Nscale secured $790M in new bank financing from ABN AMRO, DNB, Nordea, Eksfin, and SEB for its Narvik data center — the largest AI infrastructure investment in Norway — and is preparing to deliver 66,000 NVIDIA Rubin GPUs to Microsoft at a second site in Portugal.
TELUS Locks 60,000 GPUs Across Three BC Sites for Canada's $9B Sovereign AI Cluster
TELUS and the Government of Canada are building a 150 MW sovereign AI factory cluster in British Columbia powered by NVIDIA Vera Rubin, with the first facility of three coming online later this year and full capacity by 2032.
Claude Platform on AWS Goes GA: Full Feature Parity, IAM Auth, Single Invoice
Anthropic's native Claude Platform is now accessible via existing AWS accounts with IAM authentication, CloudTrail audit logging, and billing that retires against existing AWS commitments — same-day feature parity with the native API, no Bedrock required.
CopilotKit Raises $27M as AG-UI Protocol Hits Google, Microsoft, Amazon, and Oracle
The Seattle startup behind AG-UI, the open protocol for connecting AI agents to user-facing applications, closed a $27M Series A led by Glilot, NFX, and SignalFire. AG-UI has been adopted by all four hyperscalers and most major agent frameworks. CopilotKit's tools now see 4 million weekly downloads and are used in production by the majority of the Fortune 500.
Google Confirms First AI-Generated Zero-Day Exploit in the Wild — 2FA Bypass, Mass Campaign Planned
Google's Threat Intelligence Group identified a cybercrime group that used an AI model to discover and weaponize a zero-day vulnerability in a popular open-source system admin tool. The code signature is unmistakable: hallucinated CVSS score, educational docstrings, textbook Python structure. Google patched it before the mass campaign launched.
Thinking Machines Ships TML-Interaction-Small: 0.40s Latency, 77.8 FD-Bench, Doubles GPT-Realtime-2.0
Mira Murati's Thinking Machines Lab released a research preview of TML-Interaction-Small, a 276B MoE model with 12B active parameters designed for full-duplex voice and video interaction. It processes 200ms micro-turns simultaneously, achieving 0.40s turn-taking latency versus 1.18s for GPT-realtime-2.0 and 0.57s for Gemini-3.1-flash-live.
Fired for Using It, Fired for Not: China's AI Mandate Economy Is Now Measurable
In Shenzhen, the city subsidises OpenClaw installations. At Chinese tech firms, employees track daily AI automations on spreadsheets — miss too many days and face consequences. Meanwhile, UCF graduates are booing AI mentions at commencement. Two countries, same technology, opposite responses.
UCF Humanities Grads Boo AI Speaker. Jensen Huang Told CMU to Run Toward It.
On May 8, thousands of University of Central Florida humanities graduates booed a commencement speaker who called AI the next industrial revolution. Three days earlier, Jensen Huang told Carnegie Mellon grads to run, not walk, toward it. The split is now public and on camera.
NVIDIA Ships cuda-oxide v0.1.0: Write GPU Kernels in Pure Rust, No C++ Required
NVIDIA Labs releases cuda-oxide, a custom rustc backend that compiles pure Rust to CUDA PTX. Blackwell-era intrinsics included from day one. The CUDA moat just got wider.
EU Parliament Committee Bans Deepfake Fraud and Cuts AI Act Machinery Overlap in Two Amendments
The European Parliament's Internal Market and Consumer Protection Committee approved amendments adding an explicit deepfake fraud prohibition to the AI Act and creating a machinery regulation exemption that eliminates dual conformity assessments for industrial AI. Full Parliament and Council adoption pending.
HiDream O1-Image Scraps the VAE: 8B Pixel-Space Model Is Now the Top Open-Weight Text-to-Image Entry
HiDream.ai open-sourced O1-Image on May 8, removing the VAE and separate text encoder that every other diffusion model uses. The 8B unified transformer runs directly in raw pixel space and debuted at #8 on the Artificial Analysis Text-to-Image Arena — the highest-ranked open-weight model on the board. MIT licensed.
Trump's AI Security Executive Order Drops Mandatory Pre-Release Tests — Five Labs, All Voluntary
A draft White House executive order revamping federal AI cybersecurity programs will not require government approval before model deployment. Earlier proposals for mandatory pre-release testing have been shelved. All five major frontier labs now participate in voluntary government review, but Meta — the largest open-weight model provider — has no agreement in place.
Arena Adds Six Models Across Four Leaderboards in 48 Hours — Broadest Coverage Expansion of 2026
GPT-5.5-Instant enters the Text, Vision, and Document categories; ERNIE 5.1 enters Search; Gemma 4-31B and 26B-A4B enter Vision; Qwen 3.6 Max Preview and HunYuan Hy3-Preview enter Code. The two-day expansion gives the first apples-to-apples Arena comparisons for all six models.
OpenRouter Ships Unified Web Search and Fetch for Any Model — Four Engines, One API
OpenRouter's new server-side tools let any model call web search or page fetch without client-side plumbing. Exa integration costs $0.004 per search result; native provider search is included at provider pricing. The change removes the last major reason to wire search capability per-model.