OpenAI has told Elon Musk it is taking its models back. On August 28, 2026, the company notified SpaceX that Cursor access ends.
Anthropic needed about ninety minutes to answer the biggest question in developer tooling this week, and it answered with compute.
Project OT had a code name, a retreat, a playbook, and a date on the calendar. What it did not have was technology that worked.
AWS is giving away a $449-a-year training subscription to anyone who can prove they are enrolled at a university.
Britain energy regulator published a figure that reframes the entire debate about AI infrastructure in Europe.
MiniMax spent the summer doing what Chinese AI labs now treat as routine: ship a frontier-class model behind an API, then open it.
For two years ChatGPT Temporary Chat behaved like a burner phone: nothing came in, nothing went out, nothing stayed.
Max Harris listens for the hiss. The EDM producer has spent months turning his feed into an informal forensics desk.
Google DeepMind published in Nature the peer-reviewed case for a model that had already been in operational use.
A residential broadband connection began sending requests to an AI model server that should never have been reachable.
The claim that AI can do science stopped being a press release in 2026 and became a citation.
The music industry has spent nearly three years arguing that training an AI model on copyrighted songs is itself infringement. This suit changes the question.
On his 100th day in the job, Kevin Warsh spent the opening of his Jackson Hole keynote on artificial intelligence.
Poland cybersecurity agency ran a test on Meta. CERT Polska flagged 122 advertisements it had classified as fraudulent.
Roland Busch did not measure Europe regulatory lag in months. He measured it in model generations.
xAI is asking a federal judge in Denver to answer a question no court has squarely resolved: who gets to govern AI.