🤖 AI Daily · Jul 28, 2026

Kimi K3 2.8T MoE open-sourced with technical report and high-performance kernels; Google AI Overviews now appears in 43% of search results; NVIDIA, Microsoft, others launch Open Secure AI Alliance for open-source AI safety; Anthropic clarifies it never advocated banning open-weight models, supports chip export controls and safety testing; OpenAI data shows 43.5% of workers use ChatGPT for other job functions; GitHub Copilot launches 'Harness' workflow covering the full development lifecycle.

1. 1. Kimi K3 2.8T MoE Open-Sourced with Technical Report

   Moonshot AI open-sources Kimi K3: 2.8T MoE with native vision and 1M context. New architecture delivers 2.5x intelligence per compute. Includes attention kernels, MoE comms, and agent infra.

2. 2. Google AI Overviews Now Appears in 43% of Search Results

   Google AI Overviews prevalence jumps from 15% to 43% in a year. AI Mode monthly visits grow from 126M to 279M as queries shift to conversational length.

3. 3. NVIDIA, Microsoft Launch Open Secure AI Alliance

   NVIDIA, Microsoft, Hugging Face, IBM and dozens more launch Open Secure AI Alliance, building auditable AI security defenses through open-source models and tools.

4. 4. Anthropic Clarifies: Doesn't Ban Open-Weight, Supports Chip Export Controls

   Anthropic CEO clarifies it never advocated banning open-weight models. Proposes: chip export controls on China, crack down on distillation, mandatory safety testing for powerful models.

5. 5. OpenAI: 43.5% of Workers Use ChatGPT for Other Job Functions

   Analyzing 800K+ work-related ChatGPT messages, OpenAI finds 43.5% of job-specific queries involve another profession. Marketing and engineering crossover most common.

6. 6. GitHub Copilot Launches 'Harness' Workflow for Full-Stack Development

   GitHub Copilot's 'Harness' workflow enables developers to complete the entire software lifecycle—from prototyping to code review—with a single AI tool.