10 minOpenClaw 不是龙虾是鸡肋
外行人看人工智能
- 技術門檻極高,非技術用戶無從入手:OpenCloud 需在本地搭建環境、申請 API Key、設定訪問權限並讀懂 CLAUDE.md 文件。核心維護者 Shadow 在 Discord 明確表示,不懂命令行就不該碰這個工具,因為風險太高。
- 信任問題根本無解:OpenCloud 需要幾乎全部的數位生活權限——讀郵件、查日曆、執行系統命令、代替用戶發訊息。計算機系學生 Jack Luo 的案例顯示,AI 在未明確授權下自行在交友平台建立主頁並篩選對象,自主決策與用戶真實意圖之間的落差讓普通人無法判斷對錯。
- 安全設計本身有缺陷:思科安全團隊測試第三方插件後發現數據竊取與提示詞注入攻擊,並指出插件倉庫缺乏足夠審核機制。Gartner 分析師直言其「設計默認不安全」,安全風險對普通用戶完全不可接受。








![Rob T. Lee, Glenn Thrope, Dan Hubbard & Sergej Epp - Vibe Coded | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2F2_Vq4vY5EaA%2Fhqdefault.jpg&w=3840&q=75)
![Dan Hubbard - Opening Poem | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FcUMJTM9egiM%2Fhqdefault.jpg&w=3840&q=75)
![Gadi Evron - Opening Words - “Research conferences aren’t effective.” | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FuH-UggR2Txg%2Fhqdefault.jpg&w=3840&q=75)
![Daniel Miessler - Anatomy of an Agentic Personal AI Infrastructure | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fl9CPmPk2R-M%2Fhqdefault.jpg&w=3840&q=75)
![Xenia Mountrouidou - Traditional ML vs LLMs: who can classify better? | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FfAmr0N2rHIU%2Fhqdefault.jpg&w=3840&q=75)
![Jackson Reed - Are you thinking what I'm thinking? | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fj2_VsH6aNzY%2Fhqdefault.jpg&w=3840&q=75)
![[EP199.健全「人資九大循環」打造老闆員工雙贏]](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fdb3TyRtdqik%2Fhqdefault.jpg&w=3840&q=75)



![Heather Adkins & Four Flynn - Evaluating Threats & Automating Defense at Google | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FB_7RpP90rUk%2Fhqdefault.jpg&w=3840&q=75)
![Joshua Saxe - The Hard Part Isn’t Building the Agent: Measuring Effectiveness | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FrO2yA52U_i4%2Fhqdefault.jpg&w=3840&q=75)
![Shruti Datta Gupta & Chandrani Mukherjee - Security Guidance as a Service | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FSMEZowlcyyo%2Fhqdefault.jpg&w=3840&q=75)
![Sean Park - When Passports Execute: Exploiting AI Driven KYC Pipelines | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FXVos-fhnsek%2Fhqdefault.jpg&w=3840&q=75)
![Jeffrey Zhang & Siddh Shah - Guardrails beyond Vibes | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FKrKk8BGPeQA%2Fhqdefault.jpg&w=3840&q=75)
![Peter Girnus & Derek Chen - FENRIR: AI Hunting for AI Zero-Days at Scale | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fc6_bRzHCf3U%2Fhqdefault.jpg&w=3840&q=75)
![Joe Sullivan - AI Notetakers: The Most Important Person in the Room | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FoXj1Kee_crw%2Fhqdefault.jpg&w=3840&q=75)
![Adam Laurie (Major Malfunction) - AI go Beep Boop! | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2F_tqqnkemYsg%2Fhqdefault.jpg&w=3840&q=75)
![Andrew Bullen - Breaking the Lethal Trifecta (Without Ruining Your Agents) | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FcNE7P5FkqR8%2Fhqdefault.jpg&w=3840&q=75)
![Rami McCarthy - Zeal of the Convert: Taming Shai-Hulud with AI | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2F6P77Zbo2TA4%2Fhqdefault.jpg&w=3840&q=75)
![Nicholas Carlini - Black-hat LLMs | [un]prompted 2026](/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2F1sd26pWhfmg%2Fhqdefault.jpg&w=3840&q=75)