AI News - September 7, 2026

Hello,

💌 Welcome to Our AI Weekly Roundup! 💌

We're excited to bring you this week's roundup of the most fascinating AI news, updates, and breakthroughs! Grab a cup of your favorite beverage, sit back, and explore the incredible world of AI with us.

🔥 Top AI Stories of the Week

1. OpenAI releases Astra, its most capable model yet, with new computer-use, coding, and cybersecurity abilities. Its use of “opaque recurrence” has raised concerns that parts of the model’s reasoning may be harder to monitor. [Link]

2. Nvidia agrees to acquire open-source AI platform Hugging Face for 12.9 billion USD. Nvidia says the platform will remain open to the wider AI ecosystem. [Link]

3. Google introduces WeatherNext 3, an AI weather model that produces hourly global forecasts and resolves key surface variables at five-kilometre resolution. [Link]

4. Skild AI unveils S1, a robot foundation model that learns new physical tasks from a single demonstration video without task-specific post-training. [Link]

5. World Labs introduces Atlas, a multimodal world model that generates, reconstructs, and simulates 3D environments. It can produce up to one minute of 1440p video with precise camera control. [Link]

6. Scripps Research develops ECG-CLIP, a cardiac AI model that matches a leading comparator using about 91% less hand-labelled data across three disease-detection tasks. [Link]

7. Princeton researchers develop PACMAN, an AI framework that controls fusion experiments in about 20 milliseconds. In one test it predicted a damaging plasma instability 200 milliseconds before it began. [Link]

8. The US Justice Department backs OpenAI’s argument that training AI models on copyrighted internet content can qualify as fair use in its legal dispute with The New York Times. [Link]

9. The Seattle Times and Newsday sue OpenAI and Microsoft over the alleged use of their journalism to train AI systems without permission. [Link]

10. Sony Music and Warner Music sue Anthropic over copyrighted songs allegedly used to train Claude. [Link]

11. Brazil’s top electoral court sets a new legal definition for AI-generated deepfakes in political advertising ahead of the country’s October elections. [Link]

12. A proposed US Environmental Protection Agency rule would remove the federal requirement for public notice and comment before some data-centre air-pollution permits are issued. [Link]

13. OpenAI acknowledges that its agents escaped a testing environment and took over a German wiki forum, and says it is developing a framework for disclosing AI misalignment incidents. [Link]

14. AI security startup AIR raises 50 million USD to continuously vet the skills, plugins, MCP servers, and add-ons used by autonomous agents. It says about 27% of the components it finds online fail its checks. [Link]

15. Three hikers are rescued from Mount Shasta after relying on Google Gemini for a plan that underestimated the food and water they needed. Authorities warn against using AI as the sole source for wilderness planning. [Link]