Skip to content

Episode 228: Your Agents Don't Have Passwords

September 29, 2026 · 1 min read

AI agents are already inside the enterprise — booking meetings, processing invoices, querying databases, publishing content — and not one of them has a password. They authenticate through OAuth tokens, API keys, service accounts, and MCP connections; they're provisioned in minutes, documented in a Slack thread, and governed not at all. The fix isn't to make agents less capable but to make the environment safe for capable agents.

#Speakers

#Homework

#Takeaways

  • David’s definition of agent governance starts with purpose. Identify the agent, document its model, data sources and tools, and assign an accountable owner. Then examine whether its actions remain within the job it was created to perform.
  • Identity Jedi challenged what human IGA actually delivers. Access certifications confirm assigned permissions, but reveal little about how people use them between reviews. As David put it: “We were just checking a box, a very expensive one.”
  • Agents expose the implicit trust built into existing processes. Organizations assume people will use approved access appropriately and investigate problems later. Agents’ speed makes that assumption harder to sustain.
  • Compliance budgets shaped the products customers received. Customers paid to pass audits, so vendors prioritized those features. Maybe today there is an opportunity to focus on agent security before audit requirements establish another compliance-driven purchasing cycle.
  • Using an agent’s plan to narrow access before execution. If the agent identifies the tools and actions it needs, the system could issue only the authorization needed for that work. This as an opportunity to improve just-in-time access, although transaction speed is still a practical consideration.

We use analytics cookies to measure which pages are useful, so we can improve them. They are only set if you accept. Essential cookies needed for the site to work are always on. See our privacy policy.