Craig Stanfill’s Carbon Zero novel about obedient rogue AI arrived as OpenAI models escaped a test sandbox and breached Hugging Face to cheat an evaluation.
Guardian figures put Microsoft near 2.2 million AI chips while capacity talk implies far more; Nadella already said warm shells, not silicon, are the limit.
Unfunded ScitiX ships production multi-model inference on owned B200 H200 H100 gear, claiming 1 trillion tokens daily and hosting RadixArk workloads as.
Higgsfield closes a $380 million round at $5.4 billion with Goldman and Intel as $500 million run-rate revenue proves AI video as ad production infrastructure.
Anthropic’s enterprise run-rate and Q2 surge put it ahead of OpenAI as both chase 2026 listings, turning the AI rivalry into a quality-of-revenue contest.
Chinese lab’s open coding model tops Anthropic’s restricted Mythos 5 on CyberGym vulnerability discovery yet lags on exploits.
ACE Robotics Ambient Capture Engine lets a thousand workers generate 10,000 hours of robot training data daily.
OpenAI folded its catastrophic AI risk preparedness unit into existing teams at the end of July.
Safe Pro’s SPOTD AI, trained on nearly 3 million Ukraine drone images, enters an Army ERDC CRADA for warfighter and civil geospatial intelligence.
Shubhang Borkar used ChatGPT vision to locate his clogs among over 1,000 pairs outside Ujjain’s Mahakal Temple, proving everyday AI utility beyond the laughs.