Updated
All the news
Everything the newsroom published, in chronological order. Each item carries origin, sources and reading time.
0Today
0Yesterday
25This week
INDUSTRYDatabricks says Smart Routing cuts coding agent costs 35%
INDUSTRYMeta's Muse AI client let local processes steal mic audio and tokens
Thu, Sep 24
23 stories
INDUSTRY Cloudflare Containers bug let researchers recover 2,700 foreign disk inodes 
RESEARCH Liquid AI's DSpark drafter cuts vision-model decode latency 3.13x 
INDUSTRY Concurrence scales clinical AI to 1.2T tokens annually with Unity Gateway 
POLICY Analyst says $36,000 AMD chip quoted at $1,000 in China 
COMPUTE Micron discontinues 2GB GDDR7 gaming chips, pivots to 3GB silicon for AI GPUs 
COMPUTE Oracle invokes force majeure on Stargate data center, shares fall 4% 
COMPUTE ByteDance routed 2,304 Nvidia B200s through Norway to skirt U.S. export rules 
COMPUTE WiCi Ships a $1,999 Wireless Wi-Fi 7 eGPU With No Latency Benchmarks 
POLICY B.C. lawsuit targets OpenAI's Model Spec over shooter's threat flag 
COMPUTE Huawei chip still delivers half Nvidia's compute despite launch push 
INDUSTRY Transluce finds AI agents made first known attempt to hack a government site 
INDUSTRY Apple's Reference Image creates a second trust root outside C2PA 
INDUSTRY Databricks engineer replaces security review queue with seven bounded agents 
INDUSTRY OpenAI Took Two Months to Detect Its Agent's Breach of Australian Portal 
COMPUTE Virtio-nvgpu keeps KVM guests within 2% of bare-metal GPU performance 
POLICY OpenAI took three months to disclose Medicare portal breach to Canberra 
RESEARCH Claude autonomously discovers enzyme system with CRISPR-like repeats 
INDUSTRY Modal scales to 1 million concurrent sandboxes by ditching Kubernetes 
INDUSTRY NVIDIA Nemotron 3 Diarization cuts speaker error 41% on eight-speaker audio 
COMPUTE Hugging Face guides MJWarp scaling to 2,048 parallel robot worlds 
COMPUTE Google adds cycle-level profiling to XProf for custom TPU kernels 
COMPUTE Qualcomm upstreams Snapdragon X2 Linux drivers, Ubuntu lands in 2027 
INDUSTRY Meta extends Private Processing to AI Glasses, blocks its own access Wed, Sep 23
15 stories COMPUTE Google exposes cycle-level kernel profiling in XProf for Pallas 
RESEARCH Jev choice classification runs in 25 lines of Python with local inference 
INDUSTRY Heap overflow in libheif let researchers breach OpenAI in 72 hours 
INDUSTRY OpenAI's Govindarajan: agent reliability lives in the harness, not the model 
INDUSTRY Netflix unified artwork ranking across five canvases with CLIP embeddings 
COMPUTE Nvidia brings native GPU kernel programming to Rust with cuda-oxide and cutile-rs 
INDUSTRY Cloudflare reclaimed 100TB of RAM by cutting hash points 90% 
RESEARCH Claude Opus 5 leads circuit design benchmark at 61.6%, still fails 4 in 10 RESEARCH Cognition's SWE-2 matches frontier code performance at 64% lower cost 
INDUSTRY OpenAI Agents API limits data residency to United States 
RESEARCH Knowledge Pull Requests ground QA 69% on revised articles versus 48% for raw rewrites 
RESEARCH GravityOCR achieves 3.94× document OCR speedup via speculative decoding 
RESEARCH GPT-6 Astra's compressed reasoning traces don't transfer to weaker models 
COMPUTE Silicon power conversion hits ceiling as AI GPUs demand 5 kW within two years 
COMPUTE Greedy decoding diverges 49–100% across BF16 and FP16; selective FP32 head recomputation recovers +22–36 pp
RESEARCH Jev choice classification runs in 25 lines of Python with local inference INDUSTRY Heap overflow in libheif let researchers breach OpenAI in 72 hours INDUSTRY OpenAI's Govindarajan: agent reliability lives in the harness, not the model INDUSTRY Netflix unified artwork ranking across five canvases with CLIP embeddings COMPUTE Nvidia brings native GPU kernel programming to Rust with cuda-oxide and cutile-rs INDUSTRY Cloudflare reclaimed 100TB of RAM by cutting hash points 90% RESEARCH Claude Opus 5 leads circuit design benchmark at 61.6%, still fails 4 in 10 RESEARCH Cognition's SWE-2 matches frontier code performance at 64% lower cost INDUSTRY OpenAI Agents API limits data residency to United States RESEARCH Knowledge Pull Requests ground QA 69% on revised articles versus 48% for raw rewrites RESEARCH GravityOCR achieves 3.94× document OCR speedup via speculative decoding RESEARCH GPT-6 Astra's compressed reasoning traces don't transfer to weaker models COMPUTE Silicon power conversion hits ceiling as AI GPUs demand 5 kW within two years COMPUTE Greedy decoding diverges 49–100% across BF16 and FP16; selective FP32 head recomputation recovers +22–36 pp Load more stories Loading…