Daily AI news, sourcedOfficial news, first rumor reports and substantive progress. Each entry keeps the information published at that time.
- Google's Agent Compute Engine gives each Gemini agent a hardware-isolated microVM, per The Decoder
Google announced Agent Compute Engine, which runs each AI agent in its own Titanium-secured microVM, already powers Gemini, and will open for preview in coming months, The Decoder reports. Idle VMs pause automatically; Google also open-sourced Agent Substrate, a Kubernetes runtime it says can fit ten times as many agents per server, with suspend/resume under 500 ms in a 200,000-agent test.
- Google brings conversational Gemini navigation to Apple CarPlay for US drivers, per The Verge
The Verge reports that Google is bringing its Gemini assistant for navigation to Apple CarPlay, about a year after it arrived on Android Auto. Conversational Gemini in navigation is available today for US drivers, and Google says a global rollout is also happening.
- Gemini 4 Argon closes the gap with OpenAI and Anthropic but doesn't take a clear lead
The Decoder's launch-day analysis: Artificial Analysis scores Gemini 4 Argon 53 on its Intelligence Index at high reasoning — tying GPT-6 Astra and Claude Fable 5.1, but below Claude Opus 5.5 (58) and Sonnet 5.5 (56). Argon tops AutomationBench-AA (77.5%) and Text Arena, yet loses Terminal Bench 4 (57%) to Claude Sonnet 5.5 (64%), and burns about 62,000 output tokens per task versus GPT-6 Astra's 27,000, making a typical task roughly 20% pricier at list rates. The Decoder's verdict: Argon narrows the gap, but Anthropic likely still leads.
- Google replaces Gems with Skills in Gemini, joining shift to agent-ready prompts
Google is replacing Gems with Skills in Gemini chat. Skills are detailed, reusable prompts that users invoke by typing '/' or that Gemini runs automatically. The format is based on an open standard.
- Google restricts Gemini 4 Argon's first rollout to 'trusted cyber defenders', per The Verge
Google unveiled Gemini 4 Argon, its next frontier model, citing "frontier performance in complex workflows" across software engineering, enterprise work and cybersecurity defense. Initial access is restricted to a set of trusted cyber defenders while Google takes part in the U.S. government's voluntary pre-release access process, per chief AI architect Koray Kavukcuoglu. A broader rollout comes only after frontier safeguards — misuse defense, prompt-injection protection and misalignment monitoring — are hardened; no public availability date was given. Google says the model already runs in internal workflows such as large-scale codebase migrations, and published benchmark charts showing it beating rival OpenAI and Anthropic models. The announcement came one day after OpenAI's DevDay.