OpenAI / Anthropic / Chinese LLM vendors. US enterprises begin **dropping OpenAI and Anthropic for significantly cheaper Chinese AI models**, triggering a major cost and vendor-risk reckoning for Western AI providers.[1] OpenAI. OpenAI unveils **Codex Micro, a custom low‑latency edge AI chip**, alongside a new **GPT‑5.6 “Soul” variant that is dramatically more robust under adversarial prompt stress tests**, signaling a push into hardware and safer frontier models.[6] Microsoft. Microsoft rolls out **deep Copilot integration across its cloud infrastructure stack (databases, DevOps, security, Azure)**, making AI assistance available across nearly the entire Microsoft enterprise ecosystem rather than just Office-style apps.[3] Meta. Meta publishes **Llama 4 domain‑specific research showing strong results in protein structure prediction and materials science**, evidence that open‑weight models are rapidly closing the gap with proprietary systems in specialized scientific work.[3] Google DeepMind. DeepMind leadership releases a **detailed frontier‑AI risk governance framework calling for a US‑led international standards body to test the most powerful models pre‑deployment**, escalating policy pressure around safety evaluations of top‑tier systems.[5]