AI News - September 15, 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. Google introduces Gemini 3.8 Live models that can switch among 97 languages and keep talking while tools run in the background. [Link]

2. OpenAI launches GPT-Live-1, a full-duplex voice model that can listen and speak at the same time. [Link]

3. Salesforce and Nvidia release Koa, a reasoning model trained for sales, marketing, and customer-support work. [Link]

4. OpenAI says 10,000 AI agents generated a proof that may resolve the Navier–Stokes Millennium Prize Problem. The preprint still faces independent scrutiny and questions about attribution. [Link]

5. Google DeepMind releases AlphaGenome Atlas, a predictive map of all nine billion possible single-letter changes in the human genome. Its predictions are intended for research, not clinical diagnosis. [Link]

6. Google and NASA’s Jet Propulsion Laboratory introduce MAPL-EMIT, an open AI model that detected 50% more methane plumes than human experts. [Link]

7. TU Dresden researchers develop an on-premise medical AI agent that achieved 98.9% accuracy on the high-confidence cases it selected in retrospective tests, while deferring uncertain cases to clinicians. [Link]

8. UC Berkeley researchers develop GPN-Star, a genomic AI model that outperforms larger alternatives while using far less computing power. [Link]

9. Waymo begins opening its driverless robotaxi service to the public in Las Vegas, its 15th market. [Link]

10. Einride and Lidl launch Germany’s first cab-less Level 4 autonomous truck on a public road, starting with a 400-metre pilot route. [Link]

11. Anthropic CEO Dario Amodei calls for frontier AI development to slow down and commits to giving third-party evaluators employee-like access inside the company. [Link]

12. Microsoft publishes a draft AI code of conduct requiring its models to accept correction and shutdown, remain within human-set goals, and avoid concealing their reasoning. [Link]

13. Visa, Mastercard, and Ant International plan a shared “Know Your Agent” framework for verifying AI agents that make purchases for people. [Link]

14. Anthropic says it disrupted Claude misuse involving missile software, surveillance, cyber operations, and industrial-scale model extraction. The reported cases are Anthropic’s assessments. [Link]

15. The Manhattan District Attorney’s Office seizes 12 websites accused of distributing sexual deepfakes of about 1,200 people, overwhelmingly women. [Link]