Author: RadarAI Editorial
Editor: RadarAI Editorial
Last updated: 2026-11-06
Review status: Editorial review pending
Brief
速报
官方
AI动态
开源
AI security and on-device models are this week's focal points: Cloudflare warns that AI-driven vulnerability discovery will trigger a security wave lasting up to two years, while Google open-sources EmbeddingGemma 2, a 740M-parameter model that advances on-device multimodal deployment; meanwhile, Perplexity, World Labs, and other companies continue to push the boundaries of agent capabilities and real-time world models [8][10][3][11].
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Key Insights
**AI security** and **on-device models** are this week's focal points: **Cloudflare** warns that AI-driven vulnerability discovery will trigger a security wave lasting up to two years, while **Google** has open-sourced **EmbeddingGemma 2**, a **740M-parameter** model that advances on-device multimodal deployment; meanwhile, **Perplexity**, **World Labs**, and other companies continue to push the boundaries of agent capabilities and real-time world models [8][10][3][11].
## 🚀 Top Stories
- **Cloudflare CEO warns the AI vulnerability wave will last two years** [8]: AI is uncovering hidden vulnerabilities in traditional software, and six critical vulnerabilities have already been publicly disclosed.
- **Google releases the EmbeddingGemma 2 multimodal embedding model** [10]: 740M parameters, only 567MB, deployable locally on phones, and capable of handling text, images, and audio/video in a unified way.
- **Perplexity launches the Teammate agent feature** [3]: It acts as a "team member" to carry out sustained complex work and simplifies the planning workflow of Perplexity Computer.
- **World Labs releases the Frameboy real-time world model demo** [11]: Users can precisely control the camera and explore AI-generated worlds streamed on cloud GPUs at up to 15 fps.
- **Siemens Healthineers signs a cooperation agreement with Alibaba DAMO Academy** [6]: To promote non-contrast CT + AI "one-scan multi-screening" technology for early detection of highly fatal cancers such as pancreatic cancer and gastric cancer.
- **Palo Alto Networks in talks to acquire AI agent security company Clutch** [9]: For $500 million in cash plus milestone payments, focusing on the access security of agent systems.
- **webAI open-sources the TwIL-8B reasoning model** [14]: Focused on mathematics and programming, outperforming Gemma 4 31B on MATH-500 and AIME 2025.
- **Liu Zhiyuan writes to advocate building a practice-oriented AI evaluation system** [2]: He argues for incorporating efficiency and cost into evaluation and guiding AI from scale competition toward value competition.
## 🔗 Sources
[1] The 17 products most worth watching in October | Tomorrow's Product WATCHLIST — https://www.bestblogs.dev/article/4612002be6?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[2] Liu Zhiyuan writes in Guangming Daily: Establish a practice-based artificial intelligence evaluation system — https://aihot.news/items/fepqq3huguoep1lsv0gdq9xaq
[3] Perplexity launches the Teammate agent feature, capable of carrying out sustained complex work — https://aihot.news/items/ctvle8l22suzweurk164mxxe2
[4] Hideo Kojima on AI and game development: It can improve efficiency, but may deprive creation of its most valuable trial-and-error process — https://aihot.news/items/o1nfgszs5ozb8yacpngofpnpm
[5] The philosophy behind AI safety: How does effective altruism influence Silicon Valley? — https://www.bestblogs.dev/article/438e20d610?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[6] Siemens Healthineers signs a cooperation agreement with Alibaba DAMO Academy to promote AI one-scan multi-screening technology — https://aihot.news/items/ldivi5puwzc27cazvzri84y29
[7] Honor CEO Li Jian: The battle over AI entry points is not a zero-sum game over existing shares; entry sovereignty should return to users — https://aihot.news/items/x5ybg5h
AI security and on-device models are this week's focal points: Cloudflare warns that AI-driven vulnerability discovery will trigger a security wave lasting up to two years, while Google has open-sourced EmbeddingGemma 2, a 740M-parameter model that advances on-device multimodal deployment; meanwhile, Perplexity, World Labs, and other companies continue to push the boundaries of agent capabilities and real-time world models [8][10][3][11].
🚀 Top Stories
- Cloudflare CEO warns the AI vulnerability wave will last two years [8]: AI is uncovering hidden vulnerabilities in traditional software, and six critical vulnerabilities have already been publicly disclosed.
- Google releases the EmbeddingGemma 2 multimodal embedding model [10]: 740M parameters, only 567MB, deployable locally on phones, and capable of handling text, images, and audio/video in a unified way.
- Perplexity launches the Teammate agent feature [3]: It acts as a "team member" to carry out sustained complex work and simplifies the planning workflow of Perplexity Computer.
- World Labs releases the Frameboy real-time world model demo [11]: Users can precisely control the camera and explore AI-generated worlds streamed on cloud GPUs at up to 15 fps.
- Siemens Healthineers signs a cooperation agreement with Alibaba DAMO Academy [6]: To promote non-contrast CT + AI "one-scan multi-screening" technology for early detection of highly fatal cancers such as pancreatic cancer and gastric cancer.
- Palo Alto Networks in talks to acquire AI agent security company Clutch [9]: For $500 million in cash plus milestone payments, focusing on the access security of agent systems.
- webAI open-sources the TwIL-8B reasoning model [14]: Focused on mathematics and programming, outperforming Gemma 4 31B on MATH-500 and AIME 2025.
- Liu Zhiyuan writes to advocate building a practice-oriented AI evaluation system [2]: He argues for incorporating efficiency and cost into evaluation and guiding AI from scale competition toward value competition.
🔗 Sources
[1] The 17 products most worth watching in October | Tomorrow's Product WATCHLIST — https://www.bestblogs.dev/article/4612002be6?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[2] Liu Zhiyuan writes in Guangming Daily: Establish a practice-based artificial intelligence evaluation system — https://aihot.news/items/fepqq3huguoep1lsv0gdq9xaq
[3] Perplexity launches the Teammate agent feature, capable of carrying out sustained complex work — https://aihot.news/items/ctvle8l22suzweurk164mxxe2
[4] Hideo Kojima on AI and game development: It can improve efficiency, but may deprive creation of its most valuable trial-and-error process — https://aihot.news/items/o1nfgszs5ozb8yacpngofpnpm
[5] The philosophy behind AI safety: How does effective altruism influence Silicon Valley? — https://www.bestblogs.dev/article/438e20d610?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[6] Siemens Healthineers signs a cooperation agreement with Alibaba DAMO Academy to promote AI one-scan multi-screening technology — https://aihot.news/items/ldivi5puwzc27cazvzri84y29
[7] Honor CEO Li Jian: The battle over AI entry points is not a zero-sum game over existing shares; entry sovereignty should return to users — https://aihot.news/items/x5ybg5h
← Back to Updates