AI-Native Agent Security & Compliance #2: Harness, Sandbox, and Human-in-the-Loop
北卡浙大校友会与 AI智能体联盟系列活动。
Free Community Webinar. Zoom. 60 minutes including a 5-minute break.
AI-Native Agent Security & Compliance #2: Harness, Sandbox, and Human-in-the-Loop
AI-Native Agent Security & Compliance Series — Session 2.
Session 1 clarified: When chatbots fail, they usually produce a bad snippet of text; when autonomous agents fail, they send the wrong email, delete the wrong records, or pay the wrong recipient.
Formula: Agent = Model + Harness + Tools + Permissions + Data.
Session 2 digs one layer deeper: Boundaries live in the runtime, never in prompt engineering. Why clicking “Approve” can still result in an erroneous authorization.
🎟️ Registration & Zoom Access (RSVP & Zoom)
- 报名 (RSVP): https://luma.com/0t741rbo
- Zoom: https://zoom.us/j/98725004820?pwd=7wbjPDO28RbySwpc8PodMZY1dyogxZ.1
- 会议号 (Meeting ID):
987 2500 4820 - 密码 (Passcode):
18970521 - Location: Zoom (Guests may register on Luma to add to calendar and receive event reminders)
📅 Date & Time
Saturday 19 September 2026, 5:30–6:30 PM PDT
- 加州 PDT：9月19日（周六）下午 5:30 – 6:30 (California PDT: Saturday, September 19, 5:30 – 6:30 PM)
- 东部 EDT：9月19日（周六）晚上 8:30 – 9:30 (Eastern EDT: Saturday, September 19, 8:30 – 9:30 PM)
- 北京时间：9月20日（周日）上午 8:30 – 9:30 (Beijing Time: Sunday, September 20, 8:30 – 9:30 AM)
⏱️ Agenda
- 5:30–5:55 Part 1: What is a Harness (loops, sessions, sandboxes, human approvals, logging); why prompts alone cannot guard tool calls
- 5:55–6:00 Break: 5-minute intermission
- 6:00–6:25 Part 2: Containment isolation levels, outbound egress IP/domain allowlists, mandatory human confirmations for send/delete/pay/push operations, and audit logs that remain readable the next day
- 6:25–6:30 Q&A: Community interactive session
📝 Assignment
Draw your agent’s defensive perimeters on a single sheet of paper: How many walls exist between your model, tools, and production data?
🗺️ Upcoming Sessions
- Session 3 (Oct 3): Least Privilege in Production Products
- Session 4 (Oct 17): US-China-EU Regulatory Compliance Cross-Comparison
- Session 5 (Oct 31): Pre-Launch Compliance Checklist Workshop
📚 Lecture 1 Review & Archive
- Lecture 1 Replay: https://youtu.be/VxOK8cStVfA
- Playlist: https://www.youtube.com/playlist?list=PLG4UUYcE6oTs
- Lecture 1 Page: https://zjuaanc.org/posts/2026-09-01-ai-native-agent-security-compliance-lecture-1/
- Lecture 1 RSVP Archive (Do not edit): https://luma.com/fiht7lab
🛡️ Community Notice
Free community event. We will never ask for wire transfers, bank accounts, or SMS verification codes. Offensive attack methodologies will not be taught.
🤝 Organizers & Collaborators
- Host: Zhejiang University Alumni Association of North Carolina (ZJUAANC), NACUAA AI Venture & Impact Club
- Organizers: VI AI Foundation, VI AI for Good Foundation, AI Agent Alliance
- Co-organizers: NACUAA North Carolina Chapter, South Carolina Chapter, Silicon Valley Chapter, Northern California Chapter
- Collaboration: ZJU Agents
- Media Support: North Alpha Post