https://lazfore.github.io/ Recent content in Jiarui Li on Jiarui Li — Research Notes Hugo -- 0.165.0 en-us Tue, 08 Sep 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-09-08-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Tue, 08 Sep 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-09-08-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜Agent 安全契约、IPI 搜索与 Harness 优化 https://lazfore.github.io/daily-paper/2026-09-02-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Wed, 02 Sep 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-09-02-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜能力约束、长轨迹注入与 Agent 状态安全 https://lazfore.github.io/daily-paper/2026-09-01-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Tue, 01 Sep 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-09-01-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 9 项精选｜外部策略执行、长上下文注入与恶意工具泄露 https://lazfore.github.io/daily-paper/2026-08-28-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Fri, 28 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-28-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜技能自投毒、自演化 Harness 与隐私最小化 https://lazfore.github.io/daily-paper/2026-08-27-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Thu, 27 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-27-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜动作级防护、MCP 信任边界与 Harness 自优化 https://lazfore.github.io/daily-paper/2026-08-26-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Wed, 26 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-26-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜Harness 自优化、记忆注入与技能安全 https://lazfore.github.io/daily-paper/2026-08-19-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Wed, 19 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-19-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜委托权限、技能组合攻击与可验证记忆 https://lazfore.github.io/daily-paper/2026-08-18-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Tue, 18 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-18-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 9 项精选｜Harness 协同演化、Agent 恢复与技能可靠性 https://lazfore.github.io/daily-paper/2026-08-17-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Mon, 17 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-17-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 1 项精选｜Secret Egress、插件来源与状态隔离 https://lazfore.github.io/daily-paper/2026-08-14-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Fri, 14 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-14-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 9 项精选｜运行时安全契约、IPI 环境生成与 Harness 迁移 https://lazfore.github.io/daily-paper/2026-08-13-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Thu, 13 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-13-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 9 项精选｜Harness 自优化、Agent 权限与多智能体传播安全 https://lazfore.github.io/daily-paper/2026-08-12-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Wed, 12 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-12-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜安全 Harness 演化、行为基准与间接注入 https://lazfore.github.io/daily-paper/2026-08-11-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Tue, 11 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-11-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜持久载体攻击、多步注入与 Harness 演化 https://lazfore.github.io/daily-paper/2026-08-08-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Sat, 08 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-08-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜Harness 评测、自演化安全与多模态注入 https://lazfore.github.io/daily-paper/2026-08-07-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Fri, 07 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-07-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 9 项精选｜自演化 Harness、间接注入与技能隐私 https://lazfore.github.io/daily-paper/2026-08-06-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Thu, 06 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-06-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜间接注入检测、Agent 形式化验证与持久记忆安全 https://lazfore.github.io/daily-paper/2026-08-05-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Wed, 05 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-05-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 9 项精选｜可执行 Harness 学习、Agent 记忆攻击与工具安全 https://lazfore.github.io/daily-paper/2026-08-04-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Tue, 04 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-04-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜工具授权、记忆溯源与低成本 Harness https://lazfore.github.io/daily-paper/2026-08-03-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Mon, 03 Aug 2026 08:30:00 +0800 https://lazfore.github.io/daily-paper/2026-08-03-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 8 项精选｜Harness 安全、记忆回滚与多模态注入 https://lazfore.github.io/daily-paper/2026-07-31-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ Fri, 31 Jul 2026 08:33:00 +0800 https://lazfore.github.io/daily-paper/2026-07-31-%E7%83%AD%E7%82%B9%E8%AE%BA%E6%96%87%E6%8E%A8%E9%80%81/ 10 项精选｜Harness 自演化、Agent 记忆安全与语义信息流控制 https://lazfore.github.io/about/ Mon, 01 Jan 0001 00:00:00 +0000 https://lazfore.github.io/about/ <p>My name is Jiarui Li. I work at the intersection of <strong>AI security</strong>, <strong>applied cryptography</strong>, and <strong>practical systems</strong>, with a focus on <strong>LLM privacy</strong> and <strong>trustworthy AI</strong>.</p> <p>This blog is where I turn reading and prototyping into structured notes—problem definitions, threat models, techniques, and what I think still doesn’t work.</p> <p><strong>Topics I write about</strong></p> <ul> <li>Prompt privacy &amp; semantic leakage in LLM applications</li> <li>Threat modeling for enterprise LLM deployments</li> <li>Secure inference / privacy-preserving ML systems</li> <li>Integrity in federated learning</li> </ul>