AI Round-up: 28 August 2026

Anthropic opened a research preview of its Model Hardware Standard for safer AI interaction with physical equipment, while OpenAI joined Anthropic, AWS, Google, Microsoft and Oracle in calling for a global surge in cyber defence. Google DeepMind also began piloting double-blind evaluations intended to reduce bias in frontier-model comparisons.

Elsewhere, Gemini extended Omni 1.1 Flash video generation, Cohere released Parse 5, Microsoft introduced Agent Hooks and Replit added intelligent model routing. A judge blocked the Pentagon’s Anthropic blacklist, Nvidia paused new cloud-computing deals, and researchers linked Cursor’s AI coding tools to a cybercrime campaign.