Security Weekly - A CRA Resource: Applying Zero Trust Principles to Agents - Kieran Human - ASW #397

Sandboxing, least privilege, and monitoring are well-established controls in terms of the defenses they provide against unexpected and unauthorized actions. But being well-established in theory doesn't always translate to successful in practice. Kieran Human talks about some of the properties that a good sandbox should have and how monitoring creates a feedback loop to refine allow lists and access controls. In practice, the potential unpredictable behavior of an agent isn't much different from malware. We talk through some of the ways orgs can securely deploy agents without unnecessarily increasing their attack surface.

Resources
- https://www.threatlocker.com/blog/the-principle-of-least-privilege-for-ai-agents
- https://www.threatlocker.com/blog/applying-threatlocker-to-agentic-ai-tools
- https://www.threatlocker.com/blog/why-the-five-eyes-alliance-sees-zero-trust-as-the-best-defense-against-agentic-ai-threats

This interview is sponsored by ThreatLocker. Visit https://securityweekly.com/threatlocker to learn more about them!

Visit https://www.securityweekly.com/asw for all the latest episodes!

Show Notes: https://securityweekly.com/asw-397

00:00:00 Welcome to Application Security Weekly 397
00:01:57 Understanding Agent Sandbox Escapes and Good Sandbox Properties
00:06:30 Applying Zero Trust and Allow Listing to AI Agents
00:11:06 Agent Evasion, Monitoring, and Prompt Injection Risks
00:16:22 Securing Developer Agents with Governance and Least Privilege
00:23:56 Stopping Shadow AI and Refining Security Observability
00:28:43 AI Agents in Alerts, MCPs, and Malware Comparison
00:34:40 Cloudflare Spectre Attacks and Threat Model Updates
00:43:11 Copilot Introduces Flaw, AI Agents and Code Review
00:48:39 Applying OWASP Principles to Agentic AI Skills
00:59:50 Measuring AI Agent Exploitation Capabilities with Benchmarks