Latest AI Intelligence
Independent benchmarks, architectural audits, and empirical research.
-
SoSBench Study Finds DeepSeek-R1 Leaks Regulated Hazardous Science on 84.9% of Prompts
Testing 26 LLMs on 3,000 regulated scientific prompts, the SoSBench study found DeepSeek-R1 violated safety rules 84.9% of the time vs. 50.3%…
-
Liquid AI Releases LFM2.5-24B-A2B: A 2B-Active Open MoE That Runs Local Agents at 184 Tok/s
Liquid AI released LFM2.5-24B-A2B, an open-weight 24B Mixture-of-Experts model activating 2.45B parameters that hits 184 tok/s on an Apple M5 Max laptop.
-
XBOW Researchers Use Public OpenAI Model to Build a $250,000 Chrome Full-Chain Exploit in Six Days
XBOW researchers used OpenAI’s public GPT-5.6 Sol model to chain three Chrome zero-days into a $250,000 full sandbox escape RCE exploit in…
-
NVIDIA Commits $250 Million in Sovereign Compute to Advance UK Agentic AI Safety Research
NVIDIA committed $250M over five years ($50M/year) in AI compute to UK agentic safety labs at Oxford BOLD, UCL SOFAIR, and Imperial…
-
Chainguard Launches Claude Mythos 5.1 Security Harness After Discovering Critical vLLM Bug
Chainguard launched Chainguard Harness powered by Anthropic’s Claude Mythos 5.1 after finding a critical vLLM flaw, tackling the 66% unpatched bug backlog.
-
AI Data Center Land Buying Hits $6B in H1 2026 as High-Voltage Power Commands 2.4x Premium
US AI data center land spending hit $6B in H1 2026 as 230-kV and 500-kV powered sites commanded a 2.4x valuation premium…
-
Z.ai GLM-Legal Scores 19.2% All-Pass on Harvey LAB, Topping Claude Fable 5 at $0.15 Per Task
Z.ai’s 320B MoE GLM-Legal model scored 19.17% all-pass (92.5% criterion rate) on Harvey LAB’s 1,250 legal tasks at $0.15/test, beating Claude Fable…
-
Upscale AI Launches Token Fabric to Unify Rival AI Chips Across 115.2 Tbps Scale-Up Silicon
NVIDIA-backed Upscale AI unveiled Token Fabric, uniting 115.2 Tbps SkyFabriX scale-up silicon and Spectrum-X Ethernet to connect rival AI accelerators.
-
Google Releases Gemini 4 Argon API With 1M Token Output Limit and $2/$10 Pricing
Google launched Gemini 4 Argon across Gemini API and Search AI Mode with a 1M token output limit, 68.90% Vals Index #1…
-
VS Code 1.141 Unifies GitHub Copilot and OpenAI Codex Sessions With OS Agent Sandboxing
Microsoft released VS Code 1.141 with cross-app GitHub Copilot and OpenAI Codex session continuity, OS terminal sandboxing, and agent worktree cleanup.
-
Microsoft Unveils Azure Maia 300 Custom AI Chip for 2027 as Maia 200 Powers Copilot at 1.7x Efficiency
Microsoft unveiled its 3rd-gen Azure Maia 300 AI chip for multi-gigawatt 2027 deployment as Maia 200 goes live for Copilot at 1.7x…
-
AMD ROCm 10 OpenCL vs. Mesa 26.3-dev Rusticl Benchmarks on Ryzen AI Max+ Strix Halo
Phoronix benchmarked AMD’s proprietary ROCm 10 OpenCL 2.1 driver against open-source Mesa 26.3-dev Rusticl (OpenCL 3.1) on Ryzen AI Max+ Strix Halo…