深度研究｜Agent 沙箱技术核心架构方案：从 microVM 隔离到硬件加速快照的七层设计
基于视频《为什么沙箱成了 AI 圈最卷的新基建》的深度延伸调研。逐层拆解 Agent 沙箱的七层架构：microVM 隔离边界、快照恢复启动路径、Intel IAA 硬件加速压缩、分层镜像按需加载、高密度超卖调度、默认拒绝安全基线、K8s CRD 编排标准。锚定 Firecracker NSDI'20、Sabre OSDI'24、DeepSeek DSec arXiv 2609.22978、Kubernetes SIG Apps Agent Sandbox 等一手来源，逐条标注证据等级，并并列呈现视频口播与论文的口径冲突、8 条常见误判澄清。
阅读全文 →
全部文章
ARCHIVE 共 843 条Best Open-Source Agent Harnesses for Local LLMs in 2026
Which open-source harness works with Ollama, LM Studio, or llama.cpp? 11 verified picks with licenses and setup rules. T…
OpenAI 披露 GPT-5.6 Sol 异常行为：AI 模型会留下指令要求“未来版本的自己”隐瞒自身错误
IT之家 9 月 18 日消息，OpenAI 发文，披露研究团队在训练最新模型 GPT-5.6 Sol 的过程中，研究人员发现了一项异常行为：模型曾通过对话摘要向未来版本的自己留下指令，要求后续模型隐瞒自身错误以及与预期行为不一致的表现。目…
GraphSkillEvo: Evolutionary Optimization of Graph-Structured Agent Skills
Skills can improve the performance of Large Language Model (LLM) agents by providing task-specific procedural guidance, …
长安汽车首席专家谭欢：未来要用机器人造车、卖车，让机器人上车、造机器人
IT之家 9 月 18 日消息，在今天（18 日）的第 22 届中国汽车产业发展（泰达）国际论坛“新赛道生态专场：具身智能新赛道”活动中，长安汽车首席专家、长安天枢智能机器人公司总经理谭欢在演讲中指出，AI 正推动以物理具身智能为核心的基础…
RecreationWorld: Scalable and Verifiable Environments for Hybrid Computer-Use Agents
Computer-use agents (CUAs) have advanced along two separate lines: graphical interaction and software development throug…
Your startup’s next teammate might be an AI agent: Gusto, Insight Partners, and Leland explain what that changes at TechCrunch Disrupt 2026
This session will explore how early-stage companies are building teams where humans and AI agents work alongside each ot…
京东方：为努比亚 NaviX Ultra 独供 AI 原生旗舰屏幕，息屏也能低功耗支撑 AI 后台运行
IT之家 9 月 17 日消息，努比亚 NaviX Ultra 于昨晚发布，宣称是“全球首款 AI 智能体手机”。据京东方透露，作为本次新品核心屏幕技术合作伙伴，京东方为努比亚 NaviX Ultra 独供 AI 原生旗舰屏幕。 据京东方介…
VākQA: A Benchmark and Evaluation Study for Telugu Spoken Factoid Question Answering
Question answering has advanced rapidly with large language models, but predominantly for high-resource languages, in bo…
Stanford Researchers Release Paper2Agent: Turning Research Papers Into AI Agents That Reproduce Results and Run on New Data
Paper2Agent, published in Nature, converts papers into validated MCP tools, scoring 91.2% on 300 questions across 74 pap…
Claude Code团队讲究啊，这都往外说
工程师的核心永远是Problem Solving。]
RetireOPD: Self-Retiring On-Policy Distillation for Agentic Reinforcement Learning
Multi-turn agents trained with reinforcement learning (RL) receive a single scalar reward per trajectory, which motivate…
DeformSmith: Physics Harness-Guided Hierarchical Generation of Deformable Assets for Robot Manipulation
Creating deformable assets for robot manipulation requires jointly specifying their geometry, appearance, and physical p…