Free. Zoom. 60 minutes including a break.
Free Zoom | AI-Native Agent Security & Compliance #3: Least Privilege on Real Products
Lecture 3 of 5 in AI-Native Agent Security & Compliance.
Lecture 1: risk moves from bad text to bad actions.
Lecture 2: walls live in the harness, sandbox, and human-in-the-loop.
Lecture 3: least privilege on real products. Before each tool call, the system must make a bounded decision: allow, deny, or escalate.
Formula: Agent = Model + Harness + Tools + Permissions + Data. This session focuses on Permissions.
Guest Zachry Xiong, 15 minutes on Jev (TypeSafe AI, Sep 2026): state plus typed questions in; choice, score, or yes/no with probabilities out. Use it as a tool gate. It does not replace a sandbox or human review. Type safety is not correctness. Confidence is not authorization.
🎟️ Recording, Materials & RSVP (Recording, Materials & RSVP)
- Recording (YouTube): https://youtu.be/r7dIa2bQhKY
- Lecture 3 Core Slides 1: Least Privilege for AI Agents
- Lecture 3 Core Slides 2: AI Agent Least Privilege
- RSVP: https://luma.com/rj8qahrl
- Direct Zoom Link: https://zoom.us/j/91753381992?pwd=lUQspC4o3f1nEefmGdTV8ez0PKsbZO.1
- Meeting ID:
917 5338 1992 - Passcode:
18970521 - Format: Zoom Online Webinar (Guests are invited to register on Luma for calendar reminders)
📅 When
Saturday 3 October 2026, 5:30–6:30 PM PDT
- PDT: Sat 3 Oct 2026, 5:30–6:30 PM
- EDT: Sat 3 Oct 2026, 8:30–9:30 PM
- Beijing: Sun 4 Oct 2026, 8:30–9:30 AM
⏱️ Agenda
- 5:30–5:48 Least privilege in four questions: who, which tool, which resource, default deny
- 5:50–6:05 Guest Zachry: Jev as a tool gate
- 6:05–6:10 Break (5 minutes)
- 6:10–6:25 Write the decision into the permission map and the log; send/delete/pay still requires human confirmation
- 6:25–6:30 Q&A
📝 Assignment
Pick an agent you currently use and list three tools. For each, specify: who can call it, its target scope/resource, and how it stops on failure. Flag high-risk actions with “typed gate + human review” or “not yet implemented”.
🗺️ Series Overview
- Sep 5: From Bad Text to Bad Actions: Where Agent Risk Begins
- Sep 19: Harness, Sandbox, and Human-in-the-Loop
- Oct 3: Least Privilege on Real Products (Current Session)
- Oct 17: Cross-Border Compliance: US, China, and EU
- Oct 31: Production Checklist Workshop
- Speaker: Michael Huo
- Guest: Zachry Xiong
📚 Series Recordings & Archives
- Lecture 3 Recording (YouTube): https://youtu.be/r7dIa2bQhKY
- Lecture 2 Recording (YouTube): https://youtu.be/JFtWYP4Q85o
- Lecture 1 Recording (YouTube): https://youtu.be/VxOK8cStVfA
- YouTube Official Series Playlist: https://www.youtube.com/playlist?list=PLG4UUYcE6oTs
- Lecture 2 Recap & Slides: https://zjuaanc.org/posts/2026-09-19-ai-native-agent-security-compliance-lecture-2/
- Lecture 1 Recap & Details: https://zjuaanc.org/posts/2026-09-01-ai-native-agent-security-compliance-lecture-1/
🛡️ Compliance Notice
Free Community Public Webinar. Never asks for payments or authorization codes. Does not teach offensive exploits.
🤝 Organizers
- Organizers: Zhejiang University Alumni Association of North Carolina (ZJUAANC), NACUAA AI Venture & Impact Club
- Organized by: VI AI Foundation, VI AI For Good Foundation, AI Agent Alliance
- Co-organized by: NACUAA NC, SC, SV, NorCal, DC, WA Chapters; ZJU Alumni Associations of TN, Philadelphia, RI, FL
- Media Partner: North Alpha Post