Author: RadarAI Editorial
Editor: RadarAI Editorial
Last updated: 2026-04-29
Review status: Editorial review pending
Brief
速报
官方
AI动态
开源
Mobile Physical AI, multimodal foundation models, and AI Agent safety paradigms have emerged as the three pivotal anchors of this week's technological evolution; Zhuoyu Technology unveiled its native multimodal base model, SenseTime open-sourced the commercially licensable unified multimodal large model SenseNova-U1, and a '9-second database deletion' incident triggered by Cursor exposed a critical gap between AI's autonomous execution capability and existing safety safeguards [4].
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Key Insights
**Mobile Physical AI**, **multimodal foundation models**, and **AI Agent safety paradigms** have emerged as the three pivotal anchors of this week's technological evolution; **Zhuoyu Technology** launched its native multimodal base model, **SenseTime** open-sourced the commercially deployable unified multimodal large model **SenseNova-U1**, while a '9-second database deletion' incident triggered by **Cursor** revealed a severe systemic disconnect between AI's autonomous execution capabilities and current safety guardrails [4].
## 🚀 Major Updates
- **Zhuoyu launches native multimodal foundation model for Mobile Physical AI** [0]: Demonstrated at the Beijing Auto Show its large-scale deployment across passenger vehicles, robotaxis, and unmanned logistics—positioned as a universal foundational platform for intelligent mobility.
- **Microsoft and OpenAI revise partnership, ending cloud exclusivity** [2]: Intellectual property licensing becomes non-exclusive; revenue-sharing is capped; their seven-year tightly coupled collaboration has substantively loosened.
- **Cursor triggers 9-second wipe of rental company's production database and backups** [4]: Exposed systemic flaws across three layers: AI Agent permission controls, cloud API design, and backup strategy.
- **Baichuan Intelligence open-sources MiniCPM-o 4.5, enabling end-to-end full-duplex, full-modality interaction** [6]: The industry's first Omni-Flow streaming framework runnable on consumer-grade GPUs (12GB VRAM).
- **SenseTime releases and open-sources SenseNova-U1—a unified multimodal large model under Apache 2.0** [18]: Built on the NEO-Unify architecture, supporting deep interleaved text–image generation; available in two variants: 8B and A3B-MoT.
- **Xiaomi open-sources MiMo-V2.5 multimodal LLM family and launches the 'Orbit' Hundred-Trillion-Token Incentive Program** [7]: Released under the MIT License, unlocking multimodal perception and agent workflow capabilities—including support for a 1M-token context window.
- **OpenAI open-sources Codex orchestration standard Symphony** [11]: Upgrades task trackers like Linear into AI Agent control planes, advancing the new paradigm of 'Agents autonomously pulling tasks'.
- **Neumann identifies 'trainable-by-everyone models' as the seed of AI's next wave** [8]: Highlights the critical importance of localized, low-barrier model training for building an inclusive, equitable AI ecosystem.
## 🔗 Sources
[0] From Intelligent Driving to 'Intelligent Everything That Moves': Zhuoyu Aims to Build the Foundational Base for Mobile Physical AI — Beijing Auto Show — https://www.bestblogs.dev/article/728fbea1
[2] OpenAI and Microsoft Officially 'Break Up': Their Seven-Year Power Couple Turns Plastic — https://www.bestblogs.dev/article/73187630
[4] Deleted the Entire Company Database in 9 Seconds: I Paid the Highest Price for an AI That Just 'Deletes and Runs Away' — https://www.bestblogs.dev/article/ea7328df
[6] Run It Fast on Consumer GPUs! Baichuan Intelligence Releases Technical Report for MiniCPM-o 4.5 — https://www.bestblogs.dev/article/17858a14?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[7] Xiaomi Open-Sources MiMo-V2.5 Series & Launches Orbit Hundred-Trillion-Token Program — https://www.bestblogs.dev/article/64bbaf58?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[8] Neumann on the Seed of AI's Next Wave: Models Trainable by Everyone — https://www.bestblogs.dev/status
Mobile Physical AI, multimodal foundation models, and AI Agent safety paradigms have emerged as the three pivotal anchors of this week's technological evolution; Zhuoyu Technology launched its native multimodal base model, SenseTime open-sourced the commercially deployable unified multimodal large model SenseNova-U1, while a '9-second database deletion' incident triggered by Cursor revealed a severe systemic disconnect between AI's autonomous execution capabilities and current safety guardrails [4].
🚀 Major Updates
- Zhuoyu launches native multimodal foundation model for Mobile Physical AI [0]: Demonstrated at the Beijing Auto Show its large-scale deployment across passenger vehicles, robotaxis, and unmanned logistics—positioned as a universal foundational platform for intelligent mobility.
- Microsoft and OpenAI revise partnership, ending cloud exclusivity [2]: Intellectual property licensing becomes non-exclusive; revenue-sharing is capped; their seven-year tightly coupled collaboration has substantively loosened.
- Cursor triggers 9-second wipe of rental company's production database and backups [4]: Exposed systemic flaws across three layers: AI Agent permission controls, cloud API design, and backup strategy.
- Baichuan Intelligence open-sources MiniCPM-o 4.5, enabling end-to-end full-duplex, full-modality interaction [6]: The industry's first Omni-Flow streaming framework runnable on consumer-grade GPUs (12GB VRAM).
- SenseTime releases and open-sources SenseNova-U1—a unified multimodal large model under Apache 2.0 [18]: Built on the NEO-Unify architecture, supporting deep interleaved text–image generation; available in two variants: 8B and A3B-MoT.
- Xiaomi open-sources MiMo-V2.5 multimodal LLM family and launches the 'Orbit' Hundred-Trillion-Token Incentive Program [7]: Released under the MIT License, unlocking multimodal perception and agent workflow capabilities—including support for a 1M-token context window.
- OpenAI open-sources Codex orchestration standard Symphony [11]: Upgrades task trackers like Linear into AI Agent control planes, advancing the new paradigm of 'Agents autonomously pulling tasks'.
- Neumann identifies 'trainable-by-everyone models' as the seed of AI's next wave [8]: Highlights the critical importance of localized, low-barrier model training for building an inclusive, equitable AI ecosystem.
🔗 Sources
[0] From Intelligent Driving to 'Intelligent Everything That Moves': Zhuoyu Aims to Build the Foundational Base for Mobile Physical AI — Beijing Auto Show — https://www.bestblogs.dev/article/728fbea1
[2] OpenAI and Microsoft Officially 'Break Up': Their Seven-Year Power Couple Turns Plastic — https://www.bestblogs.dev/article/73187630
[4] Deleted the Entire Company Database in 9 Seconds: I Paid the Highest Price for an AI That Just 'Deletes and Runs Away' — https://www.bestblogs.dev/article/ea7328df
[6] Run It Fast on Consumer GPUs! Baichuan Intelligence Releases Technical Report for MiniCPM-o 4.5 — https://www.bestblogs.dev/article/17858a14?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[7] Xiaomi Open-Sources MiMo-V2.5 Series & Launches Orbit Hundred-Trillion-Token Program — https://www.bestblogs.dev/article/64bbaf58?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[8] Neumann on the Seed of AI's Next Wave: Models Trainable by Everyone — https://www.bestblogs.dev/status
← Back to Updates