Nvidia Chip in Russian Missile Exposes a Gap in US Sanctions
Ukrainian intelligence found an Nvidia Jetson chip inside a Russian cruise missile. The find shows where export controls on AI hardware fall short.
Chips & Hardware
AI & Society
ChatGPT Shows More Ads: What Free Users Are Actually Trading
ChatGPT is expanding ads to more countries. Which plans are affected, what OpenAI says about data, and why the ad-free choice is not entirely free.
AI & Society
OpenAI Disbands Its Team for Catastrophic AI Risks
OpenAI has dissolved its Preparedness team, which assessed bio and cyber risks from its AI models. It is the third safety team cut in two years.
Chips & Hardware
World Labs Trains Robots in Simulation: Reality Is Still the Hard Test
World Labs wants robots to learn from virtual variations of one real task. That could save hardware time, but it does not erase the real-world test.
Business & Investments
Nvidia Cuts Risk, Anthropic Grows: What It Means for the AI Bubble
Nvidia halved its guarantee for OpenAI’s data center while Anthropic nearly tripled quarterly revenue. What both moves reveal about the AI bubble debate.
Business & Investments
Apple Is Building AI for China Differently: What Alibaba’s Help Means
Apple is reportedly developing a China-specific AI model with Alibaba’s help. Regulation, control, and a different iPhone market are central to the move.
AI & Society
Local AI Instead of Cloud: Four Private Ways to Work on a Laptop
Local language models can keep sensitive text on your laptop and work offline. Four tools, their strengths, and their practical limits.
Regulation & Privacy
AI Agents in a Cyberattack: Defenders Must See the Chain
A reported attack on Taiwanese agencies shows how AI agents can link known weaknesses into an attack chain. Defenders now need to spot the connections between steps.
Regulation & Privacy
ChatGPT’s New “Computer History”: What the Mac App Actually Logs
OpenAI is replacing the screenshot-based Chronicle feature with Computer History. The Mac app logs clicks and keystrokes and stores them unencrypted.
AI Research
Why AI Models Still Misread Images
A new benchmark tests AI image understanding in its smallest parts. Even the best models stay below 60 percent accuracy.