TODAY’S RADAR · 2026-09-27

1. 🟠 Understanding the Impact of LLM Watermarking on AI Agent Behavior

56pts 70c

• 来源：HN · 得分：0.222
• 链接：https://www.lasso.security/blog/the-provenance-tax-understanding-the-impact-of-llm-watermarking-on-ai-agent-behavior

2. 🟠 Allegations of sexual harassment and rape at Bay Area AI party houses

24pts 4c

• 来源：HN · 得分：0.202
• 链接：https://www.kron4.com/news/technology-ai/report-details-allegations-of-wild-parties-sexual-harassment-and-rape-at-bay-area-ai-party-houses/

3. 🟡 AI开始研究Physical AI：FSD级团队亮出首版模型Simate-beta，空降RoboDojo

• 来源：量子位 · 得分：0.179
• 链接：https://www.qbitai.com/2026/09/498271.html

4. 🟡 在云栖大会，我终于看懂了米哈游千亿AI野心

• 来源：量子位 · 得分：0.179
• 链接：https://www.qbitai.com/2026/09/497613.html

5. 🔵 LLM Agents Can Easily Tamper With Their Own Traces

Asynchronous monitoring, incident investigations, and compliance audits primarily rely on agent trac

• 来源：arXiv · 得分：0.154
• 链接：http://arxiv.org/abs/2609.30266v1