https://blog.lynxflow.co/tags/%E5%AE%89%E5%85%A8/Recent content in 安全 on Lynx 的技术博客Hugo -- gohugo.iozh-cnMon, 31 Aug 2026 00:00:00 +0800https://blog.lynxflow.co/posts/llm-privacy-self-hosted-gateway-guide/Mon, 31 Aug 2026 00:00:00 +0800https://blog.lynxflow.co/posts/llm-privacy-self-hosted-gateway-guide/<img src="https://blog.lynxflow.co/images/llm-privacy-self-hosted-gateway-guide.png" alt="Featured image of post 别把API密钥和银行卡号贴给AI中转站：自建隐私网关的开源方案全梳理" />用第三方&quot;中转站&quot;访问 Claude、GPT、Gemini，确实便宜又方便——但你有没有意识到一件事：中转站运营者能看到你发给模型的每一个字。https://blog.lynxflow.co/posts/extracting-llm-reasoning-traces-via-tool-params/Sun, 09 Aug 2026 00:00:00 +0800https://blog.lynxflow.co/posts/extracting-llm-reasoning-traces-via-tool-params/<img src="https://blog.lynxflow.co/images/extracting-llm-reasoning-traces-via-tool-params.png?v=090500" alt="Featured image of post 关掉厂商的草稿纸，递给模型一张新的：AI 推理提取的朴素攻击法" />厂商封住了前门，但忘了窗户 大模型厂商有一套默契：原始推理过程不给你看。OpenAI、Anthropic、Google 只在 API 里返回一个&quot;总结后的思考&quot;或一段加密数据，让你能把它原样传回去让模型续接上下文，但无法读取内容。