🤖 AI Daily · Jul 8, 2026

Meta releases Muse Image & Muse Video generative models; MIRA playable multiplayer world model at 20 FPS; Rowboat open-source local-first desktop AI assistant; Grok Imagine supports 15-second video; Claude Cowork expands to mobile & web; China plans to restrict foreign access to top AI models; Microsoft replaces OpenAI/Anthropic models with in-house MAI; Apple study: single neuron bypasses LLM safety alignment; US autonomous ground vehicles deployed in Ukraine; YC CEO's 37K daily AI code claim sparks controversy.

1. 1. Meta Releases Muse Image & Muse Video

   Meta Superintelligence Labs debuts Muse Image & Muse Video, its first media generation models. Muse Image is the most advanced image generator with precise instruction following.

2. 2. MIRA: Playable Multiplayer World Model at 20 FPS

   MIRA is a playable multiplayer world model trained on 10K hours of robot data, generating game frames at 20 FPS based on button inputs.

3. 3. Rowboat: Open-Source Local-First Desktop AI Assistant

   Rowboat indexes emails, meetings, Slack into knowledge graphs with persistent context memory. Built-in email client, browser, meeting recorder & code mode.

4. 4. Grok Imagine Update: 15-Second Video Support

   xAI's Grok Imagine now supports 15-second video generation. Elon Musk says quality is incredible.

5. 5. Claude Cowork Expands to Mobile & Web

   Claude Cowork is rolling out to mobile and web with cross-device sync. Beta coming first to Max users in coming weeks.

6. 6. China Plans to Restrict Foreign Access to Top AI Models

   China plans to restrict foreign access to its most advanced AI models, consulting Alibaba, ByteDance and others. Covers both closed and open-source models.

7. 7. Microsoft Replaces OpenAI/Anthropic with In-House MAI

   Microsoft replaces OpenAI and Anthropic models in Copilot with its own MAI models to cut costs. MAI already handles tens of thousands of weekly requests in Excel and Outlook.

8. 8. Apple: Single Neuron Bypasses LLM Safety Alignment

   Apple finds safety alignment is controlled by rejection and concept neurons. Inhibiting just one rejection neuron bypasses safety in models from 1.7B to 70B parameters.

9. 9. First US Autonomous Ground Vehicles Deployed in Ukraine

   Forterra has deployed 100+ Lancer autonomous ground vehicles to Ukraine over 9 months. Based on Polaris ATV, carrying 750kg of cargo.

10. 10. YC CEO's 37K Daily AI Code Claims Spark Controversy

    YC CEO Garry Tan claims 37K lines of AI-generated code daily. Developer review reveals bloated, inefficient frontend code, sparking quality debate.