AIが期間内の動向を整理
AI News Summary for June 14, 2026 | Google strengthens on-device execution, OpenAI expands Codex task continuity
Google advanced local-friendly models and device optimization around Gemma 4. OpenAI expanded Codex Windows control and remote task continuity to keep work moving while you are away. Anthropic also published guidance that makes safety-policy updates easier to track.
SOURCE CHECK
Primary Sources 8
Primary Sources
Key Points
- 1Google improved on-device execution with Gemma 4 12B and QAT checkpoints.
- 2OpenAI expanded Codex Windows control and remote continuity to reduce interruptions.
- 3Codex also added a way to save rate-limit resets for later, increasing work flexibility.
- 4Anthropic updated Transparency Hub guidance to make Claude safety-policy changes easier to follow.
Google expands Gemma 4 for easier on-device use
Google released Gemma 4 12B as a mid-sized model that handles vision and audio together. It is designed to run more comfortably on devices with around 16GB of memory and can be tried through LM Studio and Ollama. Google also added QAT checkpoints to reduce memory use and broaden local deployment options.
OpenAI strengthens Codex task continuity
OpenAI expanded Codex with Windows Computer Use and remote control. Because you can operate Windows apps while viewing them and continue from ChatGPT on iOS or Android, it becomes easier to reduce interruptions. The ability to save rate-limit resets for later is also useful for matching AI use to real work patterns.
Google Cloud aims to speed first response to AI threats
Google Cloud Security Operations and AI Threat Defense were presented as a way to connect monitoring, detection, and response, making threat handling easier to manage. For attacks that use AI, the time between detection and action matters. For security teams, speeding up the first response is the practical value.
Anthropic improves visibility into safety-policy changes
Anthropic updated Transparency Hub guidance to show changes in Usage Policy, classifiers, and threat-model update plans. The goal is to make safety-standard changes easier to review before use. For developers and administrators, that makes deployment decisions and internal explanations easier.
Related News Items
Google / feature
Google Cloud makes AI threat response easier to manage
Google / feature
Use a high-performance multimodal model on laptops
OpenAI / feature
Let Codex watch and work on Windows tasks
Google / feature
Keep Gemma 4 efficient without losing much quality
Anthropic / policy
Track Claude safety policy changes more easily