AI Agent News
实时追踪 AI Agent 赛道的重大事件、融资动向、模型发布和技术突破
最新行业资讯
实时追踪 AI Agent 赛道的重大事件、融资动向、模型发布和技术突破
重大事件时间线
OpenClaw GitHub 爆发
OpenClaw 10 天冲上 GitHub 全球 Top 10，超越 Linux 内核 Star 增速
Meta 20亿收购 Manus
Meta 以 20 亿美元收购 Manus AI，通用 Agent 赛道正式被巨头锁定
DeepSeek-V3 开源
性价比之王，成本仅 GPT-4 的 5%
Manus 一夜爆火
全球首款通用 AI Agent 在国内社交平台引发空前关注
OpenAI Deep Research
OpenAI 推出深度研究 Agent，一键生成专业研究报告
MCP Server 破 500
MCP 生态爆发，3 个月构建 500+ Server
DeepSeek-R1 震惊全球
开源推理模型，成本仅 OpenAI 的 3%，引发全球 AI 格局震动
MCP 协议诞生
Anthropic 发布 Model Context Protocol，成为 Agent 接口事实标准
Claude Computer Use
Anthropic 让 AI 首次直接操控电脑屏幕，开创计算机使用新范式
Replit Agent 全栈自动化
自然语言到上线产品，面向非工程师
Cursor ARR 破亿
史上增长最快 SaaS，AI 编程工具新王者
Claude 3.5 登顶 SWE-bench
最强编程 AI，Bug 修复能力达到初级工程师水平
Devin 发布
全球首个自主 AI 软件工程师，能独立完成完整编程任务
Nature and Science Retract 150 AI-Generated Papers With Fabricated Data
Major scientific journals retract 150 papers found to contain AI-generated fabricated data and images. Publishers announce mandatory AI-assisted authorship disclosure and enhanced fraud detection screening.
AI Synthetic Media Detection in 2025 Elections: Global Challenges and Solutions
Election security agencies report detecting over 100,000 AI-generated synthetic media pieces targeting elections across 15 countries. Coalitions form to deploy detection tools and voter education campaigns.
Tech Giants Launch Coalition for AI-Generated Content Detection Standards
Microsoft, Google, Meta, and Adobe launch the Content Authenticity Initiative for AI, establishing watermarking and provenance standards for AI-generated images, video, and audio.
Anthropic Publishes Updated Model Spec: New Guidelines for AI Behavior
Anthropic releases comprehensive update to Claude Model Spec, detailing new guidelines for handling sensitive topics, improved calibration for confidence expressions, and enhanced corrigibility principles.
Anthropic's Mechanistic Interpretability Research Finds 'Features' in Claude's Reasoning
Anthropic published landmark interpretability research identifying thousands of 'features' — linear representations of concepts — inside Claude's neural network activations. Researchers found features corresponding to concepts like 'the Golden Gate Bridge,' 'code bugs,' and emotional states. More concerning: researchers identified features active during deceptive responses. This work brings the field closer to explaining why LLMs behave as they do, a necessary precondition for reliable AI safety guarantees.
Anthropic Achieves Breakthrough in Mechanistic Interpretability Research
Anthropic researchers publish landmark paper on mechanistic interpretability, successfully mapping how Claude represents concepts internally and identifying circuits responsible for safety behaviors.
OpenAI Establishes Safety and Security Committee, Releases Enhanced Model Security Guidelines
OpenAI has established a permanent Safety and Security Committee following months of internal and external pressure around AI safety practices. The committee has published new Model Security Guidelines covering red teaming requirements, catastrophic risk thresholds, and mandatory security reviews before major model releases. OpenAI also announced a formal vulnerability disclosure program for AI-specific security issues, with bounties up to $100,000 for critical AI safety vulnerabilities.
Anthropic Publishes Constitutional AI Safety Update: Claude 3.7 Security and Jailbreak Resistance
Anthropic has released its most comprehensive AI safety update to date, detailing Constitutional AI improvements in Claude 3.7 that reduce harmful output by 89% and jailbreak attempts by 94% compared to Claude 2. The report includes new safety benchmarks, red team findings from 200+ external researchers, and a technical specification of the Responsible Scaling Policy (RSP) thresholds that would trigger halting development of more powerful models. Anthropic also published ASL-3 requirements—the safety bar required before deploying models with potential for CBRN uplift.