Securing Non-Human Identities in the Age of Agentic AI with Sarah Cecchetti, Director of Product Management at Semperis
Show notes
This episode features Sarah Cecchetti, Director of Product Management at Semperis.
A veteran identity executive, Sarah co-founded IDPro and co-authored NIST SP 800-63-3C Digital Identity Guidelines. She previously led Amazon Cognito as Head of Product at AWS, where she also open-sourced Cedar, the policy language at the center of this conversation.
In this episode, Sarah presents her Bsides Seattle talk "Identity Crisis: IAM's Wild Ride in the AI Jungle" on why the assumptions that shaped modern identity have been overturned by the pace of agentic AI. She covers where authentication and authorization standards currently fall short for non-human identities and walks through the emerging frameworks the industry is building to fill that gap.
This episode makes the case that natural language safety instructions are not a substitute for provable, external guardrails.
Guest Bio Sarah Cecchetti is a seasoned technology executive driving product management at Semperis. At AWS, she led Amazon Cognito to triple-digit growth as Head of Product and led the open-sourcing of Cedar, a new access management language. She co-founded IDPro and co-authored NIST SP 800-63-3C Digital Identity Guidelines. Sarah has designed secure identity systems for corporate clients as well as US and Canadian governments and is recognized as a top identity professional by Okta Ventures and OWI. She’s a keynote speaker at global identity conferences like Identiverse and Authenticate.
Guest Quote “[The] average enterprise has 250,000 non-human identities, and 97% of those have excessive privilege. And 68% of organizations lack AI identity controls...The concept of excessive privilege has almost been accepted by the industry at this point. That's just the way it's done.”
Time stamps 01:45 Meet Sarah Cecchetti: Seasoned Identity Executive 02:36 Sarah’s Bsides Seattle Talk: Identity Crisis: IAM's Wild Ride in the AI Jungle 04:19 How Deepfakes Broke Biometrics 06:37 The Scale of Non-Human Identities 09:34 How NHIs Differ from Human Identities 10:38 Why FIDO Doesn't Work for AI Agents 12:19 Introducing SPIFFE and Workload Identity 15:45 How SPIFFE Works in Practice 17:34 Where AI Protocols Are Falling Short 21:12 The Problem with OAuth Client Credentials 23:18 Dynamic Registration and Database Sprawl 24:38 Client ID Metadata Documents Explained 28:43 Authentication Standards: Who Wins the Client ID Field? 30:21 Cedar: Deterministic Authorization for AI Agents 33:58 Clawdrey Hepburn: Sarah's AI Agent in Practice 40:09 Conclusion and Final Thoughts
Sponsor The HIP Podcast is brought to you by Semperis, the leader in identity-driven cyber resilience for the hybrid enterprise. Trusted by the world’s leading businesses, Semperis protects critical Active Directory and Entra ID environments from cyberattacks, ensuring rapid recovery and business continuity when every second counts. Visit semperis.com to learn more.
Links
OAuth Client ID Metadata Document