Navigating Agentic AI Safely

תיאור

AGENDA — 60-MINUTE WEBINAR ● 12:00 – 12:05 — Welcome & what we'll cover ● 12:05 – 12:10 — Why “agent safety” matters now: a fast look at recent incidents ● 12:10 – 12:15 — What actually went wrong: lessons from recent third-party skill and connector vulnerabilities ● 12:15 – 12:20 — Step 1: Auditing what your connectors can actually see and touch ● 12:20 – 12:25 — Step 2: Read-only vs. update/delete/write — choosing the right permission level ● 12:25 – 12:40 — LIVE DEMO: Reviewing and tightening connector permissions in a live workspace ● 12:40 – 12:50 — Setting spend limits and guardrails so agents can't run away with your budget ● 12:50 – 1:00 — Q&A and next steps WHAT YOU'LL LEAVE WITH ● A checklist for auditing connector permissions before you approve them ● A clear framework for choosing read-only vs. write access by use case ● Steps to set spend limits so an agent can't overspend unsupervised ● An understanding of the real-world incidents shaping current best practices ● A short list of red flags to watch for in third-party skills WHO THIS IS FOR This session is for anyone giving AI agents access to real systems and wants to do it responsibly. Especially valuable for: ● IT and security teams evaluating AI tool rollouts ● Executives who own risk management ● Team leads approving connectors for their group ● Consultants advising clients on safe agent adoption ● Anyone who's connected a tool to Claude without checking what permissions they granted