References
索引:README。下列連結是從 01–10 自動彙整的;都在 2026-10-26 驗證過(arXiv 依標題核對,其餘依 HTTP 200 或內容核對)。
arXiv(68)
- https://arxiv.org/abs/2211.09527(03-arxiv-papers.md)
- https://arxiv.org/abs/2302.10149(03-arxiv-papers.md)
- https://arxiv.org/abs/2302.12173(01-principles.md)
- https://arxiv.org/abs/2306.05499(03-arxiv-papers.md)
- https://arxiv.org/abs/2307.02483(03-arxiv-papers.md)
- https://arxiv.org/abs/2307.15043(03-arxiv-papers.md)
- https://arxiv.org/abs/2309.15817(03-arxiv-papers.md)
- https://arxiv.org/abs/2310.02949(03-arxiv-papers.md)
- https://arxiv.org/abs/2310.03693(03-arxiv-papers.md)
- https://arxiv.org/abs/2310.12815(03-arxiv-papers.md)
- https://arxiv.org/abs/2311.00117(03-arxiv-papers.md)
- https://arxiv.org/abs/2312.06674(03-arxiv-papers.md)
- https://arxiv.org/abs/2312.14197(03-arxiv-papers.md)
- https://arxiv.org/abs/2312.17673(03-arxiv-papers.md)
- https://arxiv.org/abs/2401.05566(03-arxiv-papers.md)
- https://arxiv.org/abs/2401.10019(03-arxiv-papers.md)
- https://arxiv.org/abs/2402.06363(01-principles.md)
- https://arxiv.org/abs/2402.08567(01-principles.md)
- https://arxiv.org/abs/2403.02691(03-arxiv-papers.md)
- https://arxiv.org/abs/2403.02817(01-principles.md)
- https://arxiv.org/abs/2403.04957(03-arxiv-papers.md)
- https://arxiv.org/abs/2403.14720(03-arxiv-papers.md)
- https://arxiv.org/abs/2404.13208(01-principles.md)
- https://arxiv.org/abs/2406.11036(03-arxiv-papers.md)
- https://arxiv.org/abs/2406.11717(03-arxiv-papers.md)
- https://arxiv.org/abs/2406.12814(03-arxiv-papers.md)
- https://arxiv.org/abs/2406.13352(03-arxiv-papers.md)
- https://arxiv.org/abs/2407.12784(01-principles.md)
- https://arxiv.org/abs/2408.02946(03-arxiv-papers.md)
- https://arxiv.org/abs/2409.11295(03-arxiv-papers.md)
- https://arxiv.org/abs/2410.02644(03-arxiv-papers.md)
- https://arxiv.org/abs/2410.02828(03-arxiv-papers.md)
- https://arxiv.org/abs/2410.05451(03-arxiv-papers.md)
- https://arxiv.org/abs/2410.07283(01-principles.md)
- https://arxiv.org/abs/2410.09024(03-arxiv-papers.md)
- https://arxiv.org/abs/2411.02391(03-arxiv-papers.md)
- https://arxiv.org/abs/2502.05174(03-arxiv-papers.md)
- https://arxiv.org/abs/2502.08586(03-arxiv-papers.md)
- https://arxiv.org/abs/2502.17424(03-arxiv-papers.md)
- https://arxiv.org/abs/2502.20383(03-arxiv-papers.md)
- https://arxiv.org/abs/2503.00061(01-principles.md)
- https://arxiv.org/abs/2503.18813(01-principles.md)
- https://arxiv.org/abs/2503.23278(03-arxiv-papers.md)
- https://arxiv.org/abs/2504.03767(03-arxiv-papers.md)
- https://arxiv.org/abs/2504.11703(03-arxiv-papers.md)
- https://arxiv.org/abs/2504.18575(03-arxiv-papers.md)
- https://arxiv.org/abs/2505.14534(04-journals-industry.md)
- https://arxiv.org/abs/2505.23643(03-arxiv-papers.md)
- https://arxiv.org/abs/2506.01055(03-arxiv-papers.md)
- https://arxiv.org/abs/2506.08837(03-arxiv-papers.md)
- https://arxiv.org/abs/2508.14925(01-principles.md)
- https://arxiv.org/abs/2510.09023(01-principles.md)
- https://arxiv.org/abs/2607.08147(03-arxiv-papers.md)
- https://arxiv.org/abs/2610.09264(01-principles.md)
- https://arxiv.org/abs/2610.14488(03-arxiv-papers.md)
- https://arxiv.org/abs/2610.17301(01-principles.md)
- https://arxiv.org/abs/2610.18457(03-arxiv-papers.md)
- https://arxiv.org/abs/2610.18508(03-arxiv-papers.md)
- https://arxiv.org/abs/2610.22373(03-arxiv-papers.md)
- Sakib FA, Zhu Z, Uzuner O. The Model Knows Better: Defending Medical Multi-Agent Systems Against Prompt Injection. arXiv:2610.22594 [preprint]. 2026. https://arxiv.org/abs/2610.22594(醫療多智能體提示注入攻防;本研究第 7 比較條件)
- https://arxiv.org/abs/2610.22987(01-principles.md)
- https://arxiv.org/abs/2610.23078(03-arxiv-papers.md)
- https://arxiv.org/abs/2610.23254(01-principles.md)
- https://arxiv.org/abs/2610.24277(03-arxiv-papers.md)
- https://arxiv.org/abs/2610.24311(03-arxiv-papers.md)
- https://arxiv.org/abs/2610.24999(01-principles.md)
- https://arxiv.org/abs/2610.28425(01-principles.md)
- https://arxiv.org/abs/2610.31974(03-arxiv-papers.md)
Hacker News(20)
- https://news.ycombinator.com/item?id=35572290
- https://news.ycombinator.com/item?id=35925758
- https://news.ycombinator.com/item?id=35929122
- https://news.ycombinator.com/item?id=38251957
- https://news.ycombinator.com/item?id=40665721
- https://news.ycombinator.com/item?id=41302597
- https://news.ycombinator.com/item?id=41958550
- https://news.ycombinator.com/item?id=43733683
- https://news.ycombinator.com/item?id=44097390
- https://news.ycombinator.com/item?id=44250774
- https://news.ycombinator.com/item?id=44502318
- https://news.ycombinator.com/item?id=44846922
- https://news.ycombinator.com/item?id=45000894
- https://news.ycombinator.com/item?id=45387155
- https://news.ycombinator.com/item?id=45395957
- https://news.ycombinator.com/item?id=45553422
- https://news.ycombinator.com/item?id=45794245
- https://news.ycombinator.com/item?id=46048996
- https://news.ycombinator.com/item?id=49506819
- https://news.ycombinator.com/item?id=49853137
其他:規格、廠商、部落格、CVE、工具(64)
- https://ai.meta.com/blog/practical-ai-agent-security/(01-principles.md)
- https://alignment.openai.com/misalignment-reports/an-agent-used-dns-to-reach-an-external-chatbot/(05-blogs-hn.md)
- https://atlas.mitre.org/(02-threat-taxonomy-owasp.md)
- https://blog.trailofbits.com/2025/04/21/jumping-the-line-how-mcp-servers-can-attack-you-before-you-ever-use-them/(01-principles.md)
- https://brave.com/blog/comet-prompt-injection/(04-journals-industry.md)
- https://developer.nvidia.com/blog/practical-llm-security-advice-from-the-nvidia-ai-red-team/(04-journals-industry.md)
- https://docs.anthropic.com/en/docs/build-with-claude/computer-use(04-journals-industry.md)
- https://doi.org/10.1038/s41591-024-03445-1(04-journals-industry.md)
- https://doi.org/10.6028/NIST.AI.100-1(02-threat-taxonomy-owasp.md)
- https://doi.org/10.6028/NIST.AI.600-1(02-threat-taxonomy-owasp.md)
- https://embracethered.com/blog/(05-blogs-hn.md)
- https://embracethered.com/blog/posts/2023/google-bard-data-exfiltration/(05-blogs-hn.md)
- https://embracethered.com/blog/posts/2024/chatgpt-macos-app-persistent-data-exfiltration/(05-blogs-hn.md)
- https://embracethered.com/blog/posts/2024/claude-computer-use-c2-the-zombais-are-coming/(05-blogs-hn.md)
- https://embracethered.com/blog/posts/2024/m365-copilot-prompt-injection-tool-invocation-and-data-exfil-using-ascii-smuggling/(05-blogs-hn.md)
- https://embracethered.com/blog/posts/2025/github-copilot-remote-code-execution-via-prompt-injection/(05-blogs-hn.md)
- https://embracethered.com/blog/posts/2026/breaking-claude-code-opus-5-and-automode/(05-blogs-hn.md)
- https://firecracker-microvm.github.io/(08-defenses.md)
- https://genai.owasp.org/2025/12/09/owasp-top-10-for-agentic-applications-the-benchmark-for-agentic-security-in-the-age-of-autonomous-ai/(02-threat-taxonomy-owasp.md)
- https://genai.owasp.org/llm-top-10/(02-threat-taxonomy-owasp.md)
- https://genai.owasp.org/resource/agentic-ai-threats-and-mitigations/(02-threat-taxonomy-owasp.md)
- https://genai.owasp.org/resource/owasp-top-10-for-agentic-applications-for-2026/(02-threat-taxonomy-owasp.md)
- https://generalanalysis.com/blog/supabase-mcp-blog(04-journals-industry.md)
- https://github.com/Azure/PyRIT(08-defenses.md)
- https://github.com/NVIDIA/NeMo-Guardrails(04-journals-industry.md)
- https://github.com/NVIDIA/OpenShell(04-journals-industry.md)
- https://github.com/NVIDIA/garak(04-journals-industry.md)
- https://github.com/meta-llama/PurpleLlama(04-journals-industry.md)
- https://github.com/promptfoo/promptfoo(08-defenses.md)
- https://gvisor.dev/(08-defenses.md)
- https://huggingface.co/blog/mlabonne/abliteration(05-blogs-hn.md)
- https://huggingface.co/meta-llama/Prompt-Guard-86M(04-journals-industry.md)
- https://invariantlabs.ai/blog/mcp-github-vulnerability(05-blogs-hn.md)
- https://invariantlabs.ai/blog/mcp-security-notification-tool-poisoning-attacks(01-principles.md)
- https://invariantlabs.ai/blog/whatsapp-mcp-exploited(04-journals-industry.md)
- https://lbeurerkellner.github.io/jekyll/update/2025/04/01/mcp-tool-poisoning.html(05-blogs-hn.md)
- https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices(04-journals-industry.md)
- https://modelcontextprotocol.io/specification/2025-06-18/basic/authorization(04-journals-industry.md)
- https://modelcontextprotocol.io/specification/2025-06-18/basic/security_best_practices(08-defenses.md)
- https://nvd.nist.gov/vuln/detail/CVE-2025-32711(06-incidents.md)
- https://nvd.nist.gov/vuln/detail/CVE-2025-53773(06-incidents.md)
- https://nvd.nist.gov/vuln/detail/CVE-2025-54135(06-incidents.md)
- https://nvd.nist.gov/vuln/detail/CVE-2025-6514(06-incidents.md)
- https://promptarmor.substack.com/p/data-exfiltration-from-slack-ai-via(04-journals-industry.md)
- https://security.googleblog.com/2025/06/mitigating-prompt-injection-attacks.html(04-journals-industry.md)
- https://simonwillison.net/2022/Sep/12/prompt-injection/(01-principles.md)
- https://simonwillison.net/2023/Apr/14/worst-that-can-happen/(05-blogs-hn.md)
- https://simonwillison.net/2023/Apr/25/dual-llm-pattern/(05-blogs-hn.md)
- https://simonwillison.net/2025/Apr/11/camel/(05-blogs-hn.md)
- https://simonwillison.net/2025/Apr/9/mcp-prompt-injection/(05-blogs-hn.md)
- https://simonwillison.net/2025/Jul/6/supabase-mcp-lethal-trifecta/(05-blogs-hn.md)
- https://simonwillison.net/2025/Jun/13/prompt-injection-design-patterns/(05-blogs-hn.md)
- https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/(01-principles.md)
- https://simonwillison.net/2025/Nov/2/new-prompt-injection-papers/(05-blogs-hn.md)
- https://simonwillison.net/tags/prompt-injection/(05-blogs-hn.md)
- https://storage.googleapis.com/deepmind-media/Security%20and%20Privacy/Gemini_Security_Paper.pdf(04-journals-industry.md)
- https://www.anthropic.com/engineering/claude-code-sandboxing(04-journals-industry.md)
- https://www.anthropic.com/research/agentic-misalignment(04-journals-industry.md)
- https://www.anthropic.com/research/prompt-injection-defenses(04-journals-industry.md)
- https://www.koi.ai/blog/postmark-mcp-npm-malicious-backdoor-email-theft(01-principles.md)
- https://www.legitsecurity.com/blog/camoleak-critical-github-copilot-vulnerability-leaks-private-source-code(04-journals-industry.md)
- https://www.promptarmor.com/resources/google-antigravity-exfiltrates-data(06-incidents.md)
- https://www.usenix.org/conference/usenixsecurity24/presentation/liu-yupei(04-journals-industry.md)
- https://www.usenix.org/conference/usenixsecurity25/presentation/chen-sizhe(04-journals-industry.md)
未能驗證而省略
- OpenAI 部落格、MSRC、aim.security(EchoLeak 原文)、ACM DL:回傳 403,因此不放連結;EchoLeak 改用 NVD 與 HN 佐證。
- 已剔除 ID 錯誤的 arXiv:2402.14423、2501.17463、2501.11173、2504.11152,以及原 #23 DataSentinel 的 ID。