Trust, But Verify: Human-in-the-Loop for Agents That Actually Matter
summary
Demonstrates concrete human-in-the-loop patterns: a live demo shows an interrupt pattern (back-and-forth before action), reserving a flight without a credit card then asking for out‑of‑band calendar permission, and escalating to OAuth style delegation using Auth0’s token vault to issue scoped short-lived tokens that can only post to calendar and expire. Covers CIBA push notifications and temporary cards (Stripe) as higher‑risk triggers, and explains matching approval level to operation risk with examples and a working booking + calendar integration.
AI-written summary · about
chapters (10)
- 0:00Why Agents Need Human Oversight
- 1:55The Interrupt Pattern
- 2:50Using OAuth Principles for Agents
- 3:15Demo: Booking a Flight with an Agent
- 5:59Reserving a Flight Without a Credit Card
- 6:52When the Agent Needs More Permissions
- 8:16Scoped Tokens and Secure Delegation
- 10:16Human-in-the-Loop Authorization
- 11:46The Future: Push Notifications and Temporary Credentials
- 12:54Matching Human Approval to Risk
related talks
23mTrust, But Verify: Human-in-the-Loop for Agents That Actually Matter
24mBuilding Trustworthy AI Authentication & Authorization for AI Agents
29mHow Auth0 Secures AI Authentication & Authorization for AI Agents
27mSecure "Human in the Loop" Interactions for AI Agents
19mSecuring Agents with Open Standards
01h 22mIdentity for AI Agents
35mIdentity and Access Management for the Decade Ahead
17mDon't give your agent the master key