https://lghupan.github.io/tags/mcp/ Recent content in MCP on Pan Hu's Random Thoughts Hugo -- 0.159.0 en-us Fri, 06 Mar 2026 00:00:00 +0000 https://lghupan.github.io/posts/securing-ai-agents-columbia/ Fri, 06 Mar 2026 00:00:00 +0000 https://lghupan.github.io/posts/securing-ai-agents-columbia/ <p>I had the honor of giving an <a href="https://github.com/lghupan/intro-to-agentic-security/blob/main/ai-agents-security-slides.pdf">invited guest lecture</a> at Columbia University, hosted by <a href="https://fredjiang.com/">Prof. Xiaofan (Fred) Jiang</a>, on the topic of <strong>&ldquo;Securing AI Agents.&rdquo;</strong></p> <h2 id="why-ai-agent-security-matters">Why AI Agent Security Matters</h2> <p>Autonomous agents like <a href="https://openclaw.ai/">OpenClaw</a> have the potential to transform daily work — automating complex, multi-step tasks that previously required human judgment at every turn. But with that power comes a critical question: <strong>how do we ensure it&rsquo;s safe to deploy them?</strong></p> <h2 id="what-we-covered">What We Covered</h2> <p>The lecture walked through the full threat landscape and defense stack for agentic AI systems:</p>