AI Round-up: 2 October 2026

OpenAI's model-control problems remain a focus of video analysis as dots demonstrations and an interview with its product lead explore increasingly autonomous work. Anthropic shows a designer using Claude Slides, and Google DeepMind explains watermarking from digital media to biology.

Microsoft targets conversational services with real-time AI voice models. Satlyt raises funding to run AI workloads on satellites, while Google's education programme gives teachers practical experience with classroom AI tools.