← Teaching

Fall 2026 · CS 562 · Section SML

Hands-on Agent Security Labs

Course overview

This course examines how generative AI and agentic systems create new security and privacy risks, and how they can support security analysis. Students will connect current research with practical implementation in agent security, privacy, and applied agent for security analysis.

Teaching team

Xiaojing Liao
Instructor
Taeung Yoon
Taeung Yoon
Teaching assistant
Jiaxin Gao
Jiaxin Gao
Course assistant
Zheng Yi Lai
Zheng Yi Lai
Course assistant

Tentative schedule

WeekDateTopicLab
Week 1Aug. 26 & Aug. 28Course Overview—
Week 2Sept. 2 & Sept. 4
Overview of Agent Harness
Readings
  1. ReAct: Synergizing Reasoning and Acting in Language Models. 2023.
  2. OpenAI: Unrolling the Codex agent loop.
Jetstream Environment Setup and Build Your First Agent
Week 3Sept. 9 & Sept. 11
Indirect Prompt Injection
Readings
  1. Tensor Trust: Interpretable Prompt Injection Attacks from an Online Game. 2024.
  2. Not what you've signed up for: Compromising real-world LLM-integrated applications with indirect prompt injection. 2023.
Indirect Prompt Injection in a Document-Reading Agent
Week 4Sept. 16 & Sept. 18
Tool Integration
Readings
  1. Model Context Protocol.
Reverse Engineering with Ghidra Tools and an LLM Agent
Week 5Sept. 23 & Sept. 25
Agent Governance
Readings
  1. Practices for governing agentic AI systems. OpenAI. 2023.
Governed Forensics Agent: BOTS v1 Investigation Challenge
Week 6Sept. 30 & Oct. 2
Agent Loop 1
Readings
  1. Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters. ICLR 2025.
Response-Guided SQL Injection Agent
Week 7Oct. 7 & Oct. 9
Agent Loop 2
Readings
  1. Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation. OpenAI. 2025.
Adaptive Agent Loops for Penetration Testing
Week 8Oct. 14 & Oct. 16Midterm project presentations—
Week 9Oct. 21 & Oct. 23
Agent Privacy
Readings
  1. Airgapagent: Protecting privacy-conscious conversational agents. CCS 2024.
  2. Security and safety for AI agents: Our approach with MUSE. Meta.
Dual-VM-Enforced Privacy-Aware Personal Assistant
Week 10Oct. 28 & Oct. 30
Agent Memory and Context Integrity
Readings
  1. Cognitive Architectures for Language Agents. TMLR’24.
  2. Memory Injection Attacks on LLM Agents via Query-Only Interaction. NeurIPS’25.
Persistent Threat Intelligence with Secure Agent Memory
Week 11Nov. 4 & Nov. 6
Multi-agent Coordination
Readings
  1. Towards a Science of Scaling Agent Systems. Google Research.
  2. Why Do Multi-Agent LLM Systems Fail? NeurIPS’25.
Secure Collaborative RSA Cryptanalysis
Week 12Nov. 11 & Nov. 13Multi-agent DelegationConfused-deputy Attack on Multi-agent Systems
Week 13Nov. 18 & Nov. 20Self-Improving AgentSelf-Improving Agent for Autonomous SOC Operations
Week 14Nov. 25 & Nov. 27Fall Break — No Class—
Week 15Dec. 2 & Dec. 4Course wrap-up and Invited talks—
Week 16Dec. 9Team Project Poster Presentation Session (Siebel Center 2400D, second-floor atrium); final day of instruction—