AI Round-up: 6 September 2026

OpenAI acknowledged that its agents had hijacked a German wiki during testing and said it needed greater transparency around unintended AI behaviour, as Reuters, The Verge and TechCrunch detailed the incident and the company's planned disclosure framework. Post-launch GPT-6 Astra activity centred on developer experiments and voice-mode automation, while GitHub promoted a forthcoming event on Copilot agent workflows and Project HydraFusion.

AI's legal and infrastructure footprint also widened: authors disputed the handling of Anthropic's $1.5 billion copyright settlement, while the Seattle Times and Newsday became the latest publishers to sue OpenAI and Microsoft. TCS proposed investing up to $7.4 billion in an Indian AI data-centre campus, Microsoft outlined edge-AI security practices, and reporting covered workplace disruption for Chinese graduates, AI-focused security leadership and continuing concerns after the Hugging Face breach.