CISO Series Podcast Our AI Agents Are So Safe and Reliable You Can't Buy Insurance for Them
12 snips
Sep 22, 2026 Aaron Stanley, cybersecurity executive and former dbt Labs security VP, explores the growing challenge of AI agents. They discuss agent identity and ownership, why communication is central to security leadership, the dangers of alert overload, and why insurers struggle to price AI risk. The conversation closes with identity governance and security’s opportunity to help shape AI’s future.
AI Snips
Chapters
Transcript
Episode notes
Saying No Just Hides The Agent
- Saying no to an AI agent does not remove it; it merely pushes the activity into an invisible location.
- The real governance problem begins when agents use employee logins or unmanaged API tokens.
Agents Need Accountable Identities
- Every agent ultimately needs its own identity, linked to a responsible human or organization.
- Employee credentials disappear when someone leaves; detached API keys and service identities may continue operating for years.
Classify Agents Before Granting Access
- Define what each agent is before assigning permissions, because shared, ephemeral, and employee-like agents require different controls.
- Trace who invoked an agent and limit authorization to the task being performed.