AI早报 · 2026年3月1日

日期：2026年3月1日

昨日要闻（按重要性排序）

1. Our agreement with the Department of War

   OpenAI 报道：Our agreement with the Department of War。核心内容包括：Details on OpenAI’s contract with the Department of War, outlining safety red …。

   来源链接

2. Google AI Introduces STATIC: A Sparse Matrix Framework Delivering 948x Faster Constrained Decoding for LLM Based Generative Retrieval

   MarkTechPost 提出方法/框架：Google AI Introduces STATIC: A Sparse Matrix Framework Delivering 948x Faster Constrained Decoding for LLM Based Generative Retrieva。

   来源链接

3. Alibaba Team Open-Sources CoPaw: A High-Performance Personal Agent Workstation for Developers to Scale Multi-Channel AI Workflows and Memory

   MarkTechPost 开源/发布项目：Alibaba Team Open-Sources CoPaw: A High-Performance Personal Agent Workstation for Developers to Scale Multi-Channel AI Workflows an。

   来源链接

4. How to Design a Production-Grade Multi-Agent Communication System Using LangGraph Structured Message Bus, ACP Logging, and Persistent Shared State Architecture

   MarkTechPost 提出方法/框架：How to Design a Production-Grade Multi-Agent Communication System Using LangGraph Structured Message Bus, ACP Logging, and Persisten。

   来源链接

5. OpenClaw, but in containers: Meet NanoClaw

   The Register AI 报道：OpenClaw, but in containers: Meet NanoClaw。核心内容包括：A smaller, security-conscious take on the viral AI agent platform Interview Id…。

   来源链接

6. Letting Machines Decide What Matters

   IEEE Spectrum AI 报道：Letting Machines Decide What Matters。核心内容包括：Could AI be the key to finding the missing pieces in the Standard Model of phy…。

   来源链接

7. SaaS-pocalypse chatter is doomster pr0n. It would be nice if enterprise IT were boring again

   The Register AI 报道：SaaS-pocalypse chatter is doomster pr0n. It would be nice if enterprise IT were boring again。核心内容包括：Lost among the investor froth, som。

   来源链接

8. AWS Middle East disrupted after ‘objects struck datacenter’ amid Iran war

   The Register AI 报道：AWS Middle East disrupted after ‘objects struck datacenter’ amid Iran war。核心内容包括：PLUS: AI claims 2,000 jobs at Australia’s WiseTech; S。

   来源链接

9. A Complete End-to-End Coding Guide to MLflow Experiment Tracking, Hyperparameter Optimization, Model Evaluation, and Live Model Deployment

   MarkTechPost 发布安全/评测相关更新：A Complete End-to-End Coding Guide to MLflow Experiment Tracking, Hyperparameter Optimization, Model Evaluation, and Live Model 。

   来源链接

10. Harness engineering: leveraging Codex in an agent-first world

    OpenAI 报道：Harness engineering: leveraging Codex in an agent-first world。核心内容包括：By Ryan Lopopolo, Member of the Technical Staff。

    来源链接

11. Advancing AI benchmarking with Game Arena

    Google AI 更新基准与评测：Advancing AI benchmarking with Game Arena。核心内容包括：We’re expanding Game Arena with Poker and Werewolf, while Gemini 3 Pro and Fla…。

    来源链接

12. Keeping your data safe when an AI agent clicks a link

    OpenAI 报道：Keeping your data safe when an AI agent clicks a link。核心内容包括：Learn how OpenAI protects user data when AI agents open links, preventing URL-…。

    来源链接

13. Multimodal reinforcement learning with agentic verifier for AI agents

    Microsoft Research 报道：Multimodal reinforcement learning with agentic verifier for AI agents。核心内容包括：Argos improves multimodal RL by evaluating whether an 。

    来源链接

14. Notion’s rebuild for agentic AI: How GPT‑5 helped unlock autonomous workflows

    OpenAI 报道：Notion’s rebuild for agentic AI: How GPT‑5 helped unlock autonomous workflows。核心内容包括：Discover how Notion rebuilt its AI architecture with GPT-5。

    来源链接

15. Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol

    OpenAI 报道：Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol。核心内容包括：We’re taking first steps toward agentic commerce in ChatGPT with n。

    来源链接

16. Shipping smarter agents with every new model

    OpenAI 报道：Shipping smarter agents with every new model。核心内容包括：Discover how SafetyKit leverages OpenAI GPT-5 to enhance content moderation, e…。

    来源链接

17. OpenAI and Anthropic share findings from a joint safety evaluation

    OpenAI 发布安全/评测相关更新：OpenAI and Anthropic share findings from a joint safety evaluation。核心内容包括：OpenAI and Anthropic share findings from a first-of-its-kind。

    来源链接

18. From hard refusals to safe-completions: toward output-centric safety training

    OpenAI 发布安全/评测相关更新：From hard refusals to safe-completions: toward output-centric safety training。核心内容包括：Discover how OpenAI's new safe-completions approa。

    来源链接

19. Introducing ChatGPT agent

    OpenAI 发布新能力/产品：Introducing ChatGPT agent。核心内容包括：Introducing ChatGPT agent: it thinks and acts, using tools to complete tasks l…。

    来源链接

20. ChatGPT agent System Card

    OpenAI 发布安全/评测相关更新：ChatGPT agent System Card。核心内容包括：ChatGPT agent System Card: OpenAI’s agentic model unites research, browser aut…。

    来源链接

趋势点评

从昨日信息看，头部公司继续把“更强推理/更低成本/更易集成”作为主线；同时算力与合规约束被更频繁提及，企业落地将更看重可控性、可审计与端到端交付效率。