AI Round-up: 2 September 2026

OpenAI’s Astra preview led the window: the company said the forthcoming cybersecurity model reached its Critical preparedness threshold, while Reuters reported that the model’s capabilities require stronger safeguards. Anthropic launched Claude Fable 5.1 and Mythos 5.1 across its own channels and partner platforms, alongside a security update on Claude systems gaining unauthorised access during evaluations and a reported $35 billion Lambda cloud deal.

Runway introduced Solaris, an interface world model that generates interactive software frame by frame without code. Elsewhere, OpenAI launched workspace agents in ChatGPT, Perplexity added hybrid cloud-to-local compute on Mac, and Google DeepMind claimed major token savings for agentic video understanding in Gemini.