思想与社会2025 · 分享于 Hacker News
CaMeL: Defeating Prompt Injections by Design
探索《CaMeL: Defeating Prompt Injections by Design》的原文与 Hacker News 社区讨论，查看来源并加入阅读清单。
从论文，走向讨论
看看社区如何讨论这个想法。
来自 Hacker News 的原始讨论。不同的视角、疑问，以及值得继续思考的线索。
最多展示 100 条相关讨论。论文可能出现在评论中，并非原帖投稿内容。
在讨论中发现的引用
按 HN 条目顺序展示最近的 50 条引用来源。
Yrd ·
GPT-5
2,063 赞同2,482 评论
Yrexpository ·
Supabase MCP can leak your entire SQL database
848 赞同470 评论
Yjacobr1 ·
The Policy Puppetry Attack: Novel bypass for major LLMs
313 赞同231 评论
Ypvg ·
EchoLeak – 0-Click AI Vulnerability Enabling Data Exfiltration from 365 Copilot
228 赞同86 评论
Ystuxf ·
From MCP to shell: MCP auth flaws enable RCE in Claude Code, Gemini CLI and more
148 赞同40 评论
Ynonvibecoding ·
Code execution through email: How I used Claude to hack itself
137 赞同69 评论
Ynick_wolf ·
Show HN: MCP-Shield – Detect security issues in MCP servers
134 赞同39 评论
Ythe_mitsuhiko ·
Code Mode: the better way to use MCP
84 赞同11 评论
Ytomrod ·
CaMeL: Defeating Prompt Injections by Design
71 赞同16 评论
Yhgaddipa001 ·
Launch HN: Slashy (YC S25) – AI that connects to apps and does tasks
70 赞同100 评论
Ygronky_ ·
Don't trust AI agents
51 赞同21 评论
Yedf13 ·
A GitHub Issue Title Compromised 4k Developer Machines
4 赞同0 评论
Yxnx ·
Defeating Prompt Injections by Design
2 赞同0 评论
Ytheptip ·
Defeating Prompt Injections by Design
2 赞同0 评论
Ykyrra ·
Defeating Prompt Injections by Design
1 赞同0 评论
Yredbell ·
Defeating Prompt Injections by Design
1 赞同0 评论
Yschmuhblaster ·
Implementing a (Vibed) LLM Coding Agent in Prolog
1 赞同0 评论