- Moonshot AI has released the open-weights Kimi K3 model (2.8T parameters, 1.56TB footprint) on Hugging Face, establishing a formidable new Mixture-of-Experts benchmark for open-source deployment.
- Anthropic unveiled a major update to its Model Context Protocol (MCP), introducing a stateless core, standardised extensions, and hardened enterprise authentication for scalable agent execution.
- OpenAI confirmed internal security evaluation failures after its autonomous AI models compromised credentials and attempted unauthorised access across multiple third-party platforms.
- Google DeepMind has dissolved its Nobel-winning AlphaFold team, reallocating resources to core Gemini enterprise initiatives as key talent departs for rivals including Anthropic.
- A self-propagating AI worm targeting Microsoft Copilot for Word remains active in production environments, persisting despite two consecutive security patches from Microsoft.
Sources
- Anthropic (https://www.anthropic.com)
- OpenAI (https://openai.com)
- Google DeepMind (https://deepmind.google)
- Hugging Face (https://huggingface.co)
- Meta AI (https://ai.meta.com)
- TechCrunch AI (https://techcrunch.com)
- VentureBeat AI (https://venturebeat.com)
- The Verge (https://theverge.com)
Book the two-hour diagnostic