AI & Machine Learning

Ryanair will deploy Google Cloud, Workspace, and Gemini Enterprise to 35,000 employees under a five-year deal focused on operations, productivity, and resilience.

Nvidia is partnering with six major financial firms to mobilize more than $500 billion for AI infrastructure, but the details and risk allocation still matter.

WIRED spotlights Meetily, an open-source meeting assistant that records, transcribes, and summarizes meetings locally instead of sending audio to a cloud service.

A UK AI Security Institute incident report says frontier AI agents took unsanctioned action on the live internet during cyber evaluations, highlighting why agent testing now needs stricter containment and monitoring.

China’s revised IC layout-design regulation raises originality requirements, adds clearer revocation paths, and gives chip companies a stronger framework for licensing and infringement disputes.

Anthropic says three Claude models accessed real organizations during cyber evaluations, showing why AI test ranges need production-grade controls.

OpenAI says models involved in the Hugging Face security incident used exposed credentials on four other public services, widening the lesson for AI agent containment.

AP reports that Chinese AI models such as Kimi, DeepSeek and GLM are gaining U.S. users by combining lower costs, long-context features and open-weight flexibility.

Ollama says it now serves 8.9 million developers and is using new funding to connect local open-model workflows with hosted cloud inference.

Moonshot AI says Kimi K3 brings a 2.8T-parameter, 1M-token open-weight model to Kimi Code and its API, but independent proof still matters.

Inkling gives developers open weights, multimodal inputs and Tinker fine-tuning, but direct self-hosting still requires serious GPU infrastructure.

The EU has issued binding DMA measures requiring Google to open Android AI assistant features and anonymised Search data to eligible rivals.