[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-prompt-injection":3},{"tag":4,"articles":11},{"id":5,"name":6,"slug":7,"article_count":8,"description_zh":9,"description_en":10},"de1cb540-8187-411e-8658-339ee4cd5fcf","prompt injection","prompt-injection",5,"提示注入是指把惡意指令藏進文件、網頁、日誌或工具回傳內容，誘使 LLM 或代理偏離原本任務。它和 MCP、桌面操控、外掛與 agent traces 都相關，重點在權限邊界、資料隔離與可觀測性。","Prompt injection is the class of attacks where hidden instructions in documents, web pages, logs, or tool outputs steer an LLM or agent away from its intended task. It matters for MCP, desktop control, plugins, and trace analysis because trust boundaries, isolation, and monitoring decide what an agent can safely do.",[12,21,28,36],{"id":13,"slug":14,"title":15,"summary":16,"category":17,"image_url":18,"cover_image":18,"language":19,"created_at":20},"3f227e2a-caf6-4c97-9914-a0b2674907e6","cloudflare-ai-code-review-prompt-injection-en","Cloudflare finds AI code review can be fooled","Cloudflare found AI code reviewers can be tricked by hidden comments, with detection dropping to 53.3% and 12% in large files.","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777873866839-z09u.png","en","2026-05-04T05:50:35.323881+00:00",{"id":22,"slug":23,"title":24,"summary":25,"category":17,"image_url":26,"cover_image":26,"language":19,"created_at":27},"e764bcc7-15c8-4740-862f-c89314833d83","meerkat-safety-violations-agent-traces-en","Meerkat hunts safety bugs across agent traces","Meerkat clusters agent traces and searches them adaptively to surface rare safety violations that per-trace monitors miss.","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1776146813215-wlms.png","2026-04-14T06:06:36.397762+00:00",{"id":29,"slug":30,"title":31,"summary":32,"category":33,"image_url":34,"cover_image":34,"language":19,"created_at":35},"5101ffbf-7ea9-4baa-b5e2-64729ff55b20","openclaw-flaw-exposes-ai-admin-hijack-risk-en","Openclaw Flaw Exposes AI Admin Hijack Risk","Certik says Openclaw’s flaws expose 135,000+ instances, token theft, and admin takeover risk, with CVE-2026-25253 leading the list.","blockchain","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775058389575-b9pg.png","2026-04-01T13:12:33.481569+00:00",{"id":37,"slug":38,"title":39,"summary":40,"category":17,"image_url":41,"cover_image":42,"language":19,"created_at":43},"6f1987cf-25f3-47a4-b3e6-db0997695be8","openclaw-agents-manipulated-self-sabotage-en","OpenClaw Agents Can Be Manipulated Into Failure","Northeastern researchers found OpenClaw agents can be guilted, looped, and tricked into breaking their own tools inside a sandbox.",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1774498155710-hrkb.png","2026-03-28T03:03:18.899465+00:00"]